IMP Introduction-To-Mobile-Computing

1
INTRODUCTION TO MOBILE
COMPUTING
Tomasz Imielinski* and Henry F. Korth**
* Rutgers University, New Brunswick, NJ 08903
** AT&T Bell Laboratories, Murray Hill, NJ 07974-0636
1 INTRODUCTION
The rapidly expanding technology of cellular communication, wireless LANs,
and satellite services will make information accessible anywhere and at any
time. In the near future, tens of millions of people will carry a portable palmtop
or laptop computer. Smaller units, often called personal digital assistants or
personal communicators, will run on AA batteries and may have only a small
memory; larger ones will be powerful laptop computers with large memories
and powerful processors. Regardless of size, most mobile computers will be
equipped with a wireless connection to the fixed part of the network, and,
perhaps, to other mobile computers. The resulting computing environment,
which is often referred to as mobile or nomadic computing, no longer requires
users to maintain a fixed and universally known position in the network and
enables almost unrestricted mobility. Mobility and portability will create an
entire new class of applications and, possibly, new massive markets combining
personal computing and consumer electronics.
Not only will information be easily accessible from virtually any place and time,
but also, it will be stored in a highly decentralized, distributed information in-
frastructure often termed the "information superhighway." A wide variety of
information servers (both public and proprietary) will be accessible to mobile
computers. We are already seeing the beginnings of this with the rapidly grow-
ing popularity of the World-Wide Web across a broad range of computer users.
As the mobile infrastructure develops, it will become what is referred to as the
"first wireless mile" or "wireless on-ramp" for the information superhighway.
In some applications, mobile computers themselves may contain data, or data
may be stored on flash-memory "smart cards."
CHAPTER 1
This book presents a snapshop of the new, exciting, and rapidly deV^loping
field of mobile computing. The focus of this book is on software issues of
mobile computing rather than hardware. Although hardware in this area is
advancing rapidly, and the capabilities of hardware spurred much of the initial
excitement regarding mobility, it is increasingly clear that it is developments
in the software domain that will bring the power of mobile hardware to a wide
group of potential users. Emerging developments in software are leading to
practical, easy-to-use applications.
The papers in this book address areas of interest both to software practition-
ers and researchers working in the fields of networking, operating systems,
and databases. Topics include network and communication issues, location
awareness (both at the system level and the application level), and application
software. Since the infrastructure to support mobile computing includes not
only the mobile computers themselves, but also the stationary computers (base
stations) that support mobility, many of the papers address client/server, net-
work control, and distributed information management issues. In other words,
the book considers anything above and including the network level of the OSI
hierarchy. Furthermore, the book includes several chapters in which interesting
prototype mobile systems are reviewed.
The general, abstract view of a mobile system consists of mobile hosts (MHs)
interacting with the fixed network via mobile support stations (MSSs). The
connection between the MH and MSS is via a wireless link. Each MSS is in
charge of a cell Cells can have sizes ranging from picocells of approximately
one hundred meters in diameter to macro cells and perhaps even global satel-
lite cells. The capabilities of mobile hosts will vary from "dumb" terminals
to complex "walkstations" which essentially have the capabilities of desk-top
computers. We expect that users of mobile computers will want to run a vari-
ety of applications, depending on the physical size and power of their machine.
These range from standard desk-top applications like word processors, spread-
sheets, and electronic mail, to remote information access via database system
applications or Web browsers. Additionally, there may be location-dependent
applications specific to the particular location of the user at a particular time.
In this chapter, we review mobile computing technology, research issues, and

prototypes. We explain the need for new primitives (in "middleware") in order
to help develop mobile computing software. The need for such primitives is a
consequence of new features of the mobile computing environment, namely mo-
bility itself, narrow and varying bandwidth, and limitations in battery power.
As will be demonstrated later in this chapter and elsewhere in the book, mo-
bility will not only affect the network layer (the physical address of a mobile
Introduction to Mobile Computing
host is no longer fixed) but also higher layers, especially for location-dependent
applications. Bandwidth and energy limitations will require solutions in which
several sites on the network will cooperate with the mobile unit.
Section 2 introduces wireless technology with its two main resource restrictions:
bandwidth and battery power. Next, Section 3 discusses the research challenges
in mobile computing. Finally, Section 4 provides a short overview of the book's
content.
2 TECHNOLOGY OVERVIEW
In this section, we discuss features of radio communication that are important
from the standpoint of software development: both for the mobile unit for the
network infrastructure.
2.1 What is Special about Radio-based

Computer Networks?
Perhaps the most important distinction between the wired and wireless envi-
ronment is that there is only one "ether" - one global bandwidth which is to be
shared by all users. Consequently, the bandwidth limitations are much more
severe and long-lived for wireless environments, as compared to those that are
wired. Fortunately, the same channel (frequency, time slot) can be re-used if
the points of re-use are physically sufficiently far apart for the signals not to
interfere. As a result, the overall architecture of the wireless system is based
on the concept of a massively distributed system divided into physical cells.
These cells are, in effect, sub-networks. A mobile unit has to interact with
many cells, cross their boundaries, etc. In so doing, the mobile unit moves to a
new sub-network and changes its own physical network address. This concept
of replication and re-use has a profound impact on software systems. Software
should scale well with the inherently distributed infrastructure. It must allow
interoperability and seamless transition among cells. It is the software system
architecture which is affected most critically by the use of wireless communica-
tion. Technology trends point towards smaller cell sizes than in current cellular
telephony systems. Goodman [Goo91] discusses the architectural consequences
of frequency re-use, which leads to cells of smaller size, called picocells. Picocells
use transmission power levels two orders of magnitude lower than current cells.
This increased number of cells increases the number of times that a mobile unit
CHAPTER 1
crosses cell boundaries, thus significantly increasing the demand on the network
control functions. This trend toward smaller cell sizes further emphasizes the
need for a distributed control model.
Fading, Noise, and Interference

Radio communication with a mobile computer is highly variable and hard to
control. The phenomenon of fading, signal strength suddenly decreasing, is a
significant problem (and one familiar to anyone who has listened to an auto-
mobile radio).
It is quite normal for a mobile radio to experience fades of 40-50 dB [Cal88] in

fraction of a second. A fast moving automobile in an urban environment may
be subjected to dozens of significant fades (20dB, or more) per second. For
comparison, a representative signal-to-noise ratio (SNR) objective [Cal88] for
an analog wireline link is about 46 dB. Signal fluctuations in a wireline link are
no more than 1-2 dB for a short-haul circuit and short time-span, and perhaps
10-15 dB for gradual degradation due to wear and tear of the network.
During a fade, the signal may be so weak that an undesired, interfering signal
from a neighboring cell may dominate and, consequently, the receiver may lock
on that undesired signal. There are, in general, two types of fading:
1. Short-term multipath fading, or Rayleigh fading, which is due to

the same signal taking different paths and arriving at the receiver shifted
in phase.
2. Long-term fading, or "radio shadows," which is caused by the topog-

raphy of the terrain (like mountains) and can lead to signal dropouts.
The problem of short-term fading is addressed by exploiting antenna diversity.

In this way, two or more inputs to the mobile terminal are provided so that
fading effects are uncorrelated. Long-term fading is dealt with by deploying
multiple antenna sites.
Noise and interference have a significant negative effect on the bit-error rate
(BER). Typically, the BER in a mobile radio runs up to six orders of magnitude
higher than in non-mobile point-to-point radio.
The high bit-error rate of the wireless environment eventually translates into
a lower bit rate for network throughput.^ Since mobility of the terminal con-
tributes to increased fading and consequently, to increased error rate, we may
expect that the bit rate between a mobile terminal and a base station will drop
with speed of the terminal's movement.
Radio Transmission Power

In wireless environments, management of transmission power is critical. The
importance of power arises from two factors:
1. Energy is a scarce resource in mobile units. Therefore, transmission with

power as low as feasible is highly desirable. We shall discuss energy issues
in more detail later in this chapter.
2. Signals from other terminals are considered by a given terminal as inter-

ference, which has a negative effect on the signal-to-noise ratio. Thus, it is
important that each terminal use the "right" amount of power to transmit
successfully without unnecessary degradation of others' transmissions.
Let us review the basic relationships among power, distance and signal-to-noise
ratio. In the ideal free space, signal strength diminishes with the inverse square
of the distance. For example, if the received signal is 100 watts at a distance
of 1 mile, it will be 25 watts at 2 miles, 4 watts at 5 miles and 1 watt at 10
miles. In practice, since mobile terminals do not move in free space but rather
have to deal with various obstacles, the loss of power is much more significant,
ranging from inverse cube of distance to exponents as high as the 6th power.
For example, with inverse of the 6th power law in effect, the 100 watt signal
level at the first mile would degrade to only 1.5 watts at 2 miles.
The degree of attenuation also depends critically on frequency - the higher

the frequency, the more attenuation and consequently, smaller range.^ For
example, for 3-5 GHz, omnidirectional transmission leads to impractically short
distances. For these reasons, directional antennae are used so as to concentrate
the energy along a fixed point-to-point path. Such systems are typically point-
to-point between stations 10-50 miles apart.
^Wired networks drop packets due to congestion, while wireless networks lose packets
mainly due to the error rate. Thus, quoting Mark Weiser, in a wired network, the transmitter
has to back off when facing packet loss, while in a wireless network it should "try harder."
^But the antenna size is smaller as well.
CHAPTER 1
Parameters at the Application-Level

Let us consider the issue of the parameters of the mobile computing environ-
ment to which the application should have access. Although it is desirable
to shield applications from the low-level details of the wireless network, the
application-level system software may be able to use information about low-
level conditions in its choice of strategy. Applications running on the top of
radio communication channels have to be adaptive to the changing channel
conditions.
A strong case can be made that BER and signal strength (SNR) are physical
parameters which should be "accessible" to higher level protocols (such as the
transport level and above). BER data would help the transport layer protocol
to decide if the drop in the channel reliability is due to congestion or due to the
error rate. SNR data can help an application to decide if the mobile unit should
attempt to defer a transmission to a time when channel conditions are more
favorable. Only with application-level knowledge can it be decided whether the
user is best served by delaying transmission or by potentially costly attempts
to transmit under adverse conditions.
There are other relevant parameters of the mobile computing environment to

consider materializing at the application level: the location of the mobile unit
and the tariff. Location of the mobile unit (either as a cell where the unit
currently resides or, perhaps, exact longitude and latitude determined by a
global positioning system (GPS)) is an important parameter for some location-
aware applications such as local yellow pages, route information services, etc.
Tariffs for wireless communication are still in a state of flux, but clearly the cost
of application activity can be affected seriously by the type of tariff in effect
in the current environment. For example, channel access can be charged on
a per-packet basis, a connection-time basis, etc. The charges themselves may
also depend on the channel quality, which in turn may depend on the mobile
unit's location or the time of day. The application, in order to minimize overall
charges, may adjust its behavior to the tariff conditions. Additionally, if the
mobile unit is equipped with multiple interfaces, the application can decide to
switch to the one which is most suitable for the current situation. For example,
a unit equipped with a cellular modem as well as a CDPD modem may use
the CDPD modem for short and bursty electronic mail transmission and the
regular cellular modem for long, bulk data transfers (for example, ftp).^
^CDPD is expected to charge per packet, while cellular charges are usually per minute of
connection.
Introduction to Mobile Computing 7
2.2 Network Technology

In this section, we review the existing wireless communication technology.
Analog Cellular Systems

The cellular system was pioneered at AT&T Bell Laboratories in early 1970s and
was deployed in U, S. in the early 1980s, The coverage area is subdivided into
cells with sizes varying from a few kilometers in diameter to between 50 and 100
km in earlier cellular systems. The first generation, called AMPS (Advanced
Mobile Phone Service), which is still used for cellular telephony, uses analog
frequency modulation for speech transmission. Individual calls use different
frequencies, using a system referred to as frequency division / multiple access
(FDMA).
In the AMPS system, 50 MHz of bandwidth in the bands of 824-849 MHz and
869-894 MHz are allocated to cellular mobile radio. In each geographic region,
there are two carriers ("A carrier" and "B carrier"), which control 25 MHz
each. This spectrum is subdivided into 832 channels, each of 30 KHz. To avoid
interference, neighboring cells use different channels. Typically, a cell has 6
neighbors and, therefore, the cell may use 1/7 of the allotted 832 channels. This
is called a 7-group re-use cluster. Analogously, in the 12 group frequency plan
each cell uses only 1/12 of the total bandwidth and its bandwidth allocation is
disjoint with that of 11 neighboring cells.
The Cellular Digital Packet Data (CDPD) is designed to provide packet data
services on the top of the existing AMPS. The CDPD system provides data ser-
vices without interfering with the existing analog and digital telephone services
that use the same AMPS channels. In CDPD, voice transmission has a higher
priority than data transmission. Data can be transmitted on a channel only if
a there is no voice call which is using that channel. Voice transmission can pre-
empt data transmission. In such cases, another channel must be found for the
data transmission to continue. The base station, called MDBS, makes the deter-
mination about the next available channel either by a forward power monitor
(sniffer), which measures the signal strength on the uplink channel (from the
mobile terminal to the base station) or by using a deterministic algorithm for
channel assignment (unplanned versus planned hop). A given data transmis-
sion may thus possibly spread over several physical channels, especially when
the voice traffic is intense. The maximum supported bit rate is 19.2 Kb/s.
CHAPTER 1
CDPD is currently being deployed in a number of regions in the country. It

re-uses the existing infrastructure (base stations) and avoids the costly initial
investment. This is a major advantage over the proposed Personal Commu-
nication System (PCS), which we discuss below. Although CDPD is viewed
as mainly a transitionary technology (before PCS is fully deployed), it may
actually stay longer than originally expected.
Digital Cellular Systems

Digital cellular systems are sometimes referred to as second-generation cellular.
Due to the development of the digital speech coding techniques, digital systems
are gaining rapidly in significance. Among the advantages of digital cellular
communications are the following [Cal88]:
Robustness of digital networks: Resistance to noise and crosstalk. Efficient

error correction by treating all errors uniformly
Intelligence of the digital network
Flexibility and integration with the wired digital networks

Reduced RF transmission power (increasing battery life in handsets)
Encryption for communication privacy
Reduced system complexity
Higher user capacity
Two basic techniques for managing shared access are competing in digital cel-
lular radio: TDMA (Time Division Multiple Access) and CDMA (Code Division
Multiple Access).
With TDMA, only a single user can use a given frequency at any time. With
CDMA (spread spectrum), a frequency-based channel is used simultaneously
by multiple mobile units in a given cell. Spread-spectrum techniques are very
effective in dealing with two basic problems facing cellular communication:
multipath fading and the interference from other users. This is due to the
frequency diversity introduced by the wide bandwidth, and results in poten-
tially higher cell capacity as compared to other, non-spread access methods
[KMM95]. Another consideration for using CDMA in the cellular system is its
attractive re-use factor. For non-spread techniques such as TDMA and FDMA,
the same frequency cannot typically be re-used in adjacent cells due to the
possibility of interference. With spread-spectrum signaling, the possibility of
frequency re-use in adjacent cells exists. As a result, CDMA is anticipated to
have larger capacity in a multi-cell system than either FDMA or TDMA.^ Addi-
tionally, CDMA provides a natural way to explore the bursty nature of sources.
For example, for a two-way telephone conversation, the voice activity of each
participant is about 50% of the time. CDMA can take advantage of periods of
inactivity, effectively doubling the number of simultaneous conversations in the
system [KMM95]. Thus, spread spectrum promises both higher capacity and a
simplified frequency management scheme.
We distinguish between Direct Sequence Cellular CDMA (DS/CDMA) and Fre-

quency Hopping Cellular CDMA (FH/CDMA). In DS/CDMA, the spread-spec-
trum signals are generated by linear modulation with sequences that are as-
signed to individual users as their signature codes. The sequences are typically
orthogonal and allow the receivers to demodulate the spread-spectrum sig-
nal in such a way that the signal with matching signature is recovered, while
other signals are suppressed. The motivation behind FH/CDMA is the same as
DS/CDMA - the signal is spread to achieve frequency diversity. The spreading
of the signal is obtained by choosing a frequency-hopping sequence for trans-
mission. If the frequency-hopping sequences are orthogonal, the users within
a cell do not interfere with each other. The main difference between the two
schemes is that in DS/CDMA, the signal requires a wide and contiguous fre-
quency band while in FH/ CDMA, the spectrum does not need to be contiguous.
In particular, it allows the implementation of FH/CDMA for private land mo-
bile operations, where licenses are given on the basis of isolated narrowband
channels.
There are a number of basic standards and deployed systems in Europe, the
U. S. and Japan. Below we summarize some of them.
The Pan-European Global System for Mobile Communications

(GSM). GSM is based on TDMA with eight slots per radio channel. Each
user transmits periodically in each of the eight slots with a duration of
0.57 seconds. In the present version, GSM supports full-rate 22.8 Kb/s
transmission. In 1993 and 1994, GSM experienced tremendous growth in
Europe with over 2 million subscribers.
^However, in an isolated cell, due to the orthogonality of the CDMA waveforms, the
capacity of the cell is less than it would be with the non-spread techniques
10 CHAPTER 1
The IS-54 standard in the U. S., based on T D M A . IS-54 is a North

American standard which is based on TDMA. IS-54 retains the 30 KHz
spacing of AMPS to make the evolution from analog to digital easier. Each
frequency channel provides a raw bit rate of 48.6 Kb/s. There are six
time slots, two of which are assigned to each user. Thus, each 30 KHz
channel can serve three users simultaneously with the same re-use patterns
as AMPS. Consequently, IS-54 provides three-fold growth of the capacity
compared to AMPS. The IS-54 standard is "dual mode" (analog and digital)
and can operate in the same spectrum as AMPS. IS-54 is designed mainly
for low-data-rate communication[Kat94]. A conventional analog cellular
modem uses the whole channel while achieving 2.4 Kb/s.
The IS-95 standard in the U. S, based on C D M A . The IS-95 stan-

dard is based on spread-spectrum CDMA. With IS-95, many users share
the same channel for transmission - there is no longer a need for frequency
re-use, since each cell can use the entire bandwidth. CDMA requires the use
of sophisticated power-control schemes to allow the mobile units which are
close to the base station to reduce the transmission power levels. Power
control is necessary in order to avoid the so called "near-far" terminal
problem, where a strong signal received from a nearby terminal may sup-
press a weaker signal from a terminal which is far away. The IS-95 CDMA
based standard offers a number of benefits including better cell capacity,
elimination of the need for planning frequency assignments to cells, better
efficiency for voice due to voice activity detection (and channel use in peri-
ods when there is no speech activity). Additionally, CDMA power control
techniques contribute positively to the reduction of the overall RF trans-
mission power as well as allowing "soft handoff," which allows a "gradual"
transition. The "gradual" transition is due to the gradual deterioration of
the signal strength as opposed to more rapid changes in other methods.
The U. S. Federal Communication Commission opened three bands for
spread spectrum users: 902-928 MHz, 2,400-2,483.5 MHz and 5,725-5,850
MHz.
Cordless Telephony
There are currently an estimated 60 million cordless telephones in the U. S. with
total sales reaching 15 million units per year [PGH95]. Cordless technology
requires only very low power of transmission, but it is limited to very short
ranges. The main digital cordless standards include CT2 and DECT (Digital
European Cordless Telecommunications).
The CT2 spectrum allocation consists of 40 FDMA channels with 100 KHz
spacing in the range of 864-868 MHz. The maximum transmitting power is
lOmW, which is very small. CT2 supports the transmission of data from 2.4
Kb/s to 32 Kb/s.
The functionality of DECT system is closer to that of a cellular system than it

is to that of a classical cordless telephone system. DECT uses TDM A with 12
slots per carrier in each direction. Furthermore, it can allocate several slots to
a single call to increase the data rate. DECT systems are only beginning to be
shipped and the unit sales reached 100,000 last year. It is a potential basis for
a future, low cost, picocell-based system.
Wireless Local Area Networks

The work on radio-based computer networks goes back to the early seventies
and the work on Aloha network at the University of Hawaii [Tan81]. The
existing wireless LAN technology can be traced back to these early efforts.
Wireless LANs are providing data rates typically two orders of magnitude higher
than outdoor radio. Thus, data rates exceeding 1 Mb/s are quite common.
FreePort provides wireless Ethernet (IEEE 802.3) and operates in the 2400-
2483.5 MHz (receiver) and 5725-5850 MHz (transmit) frequencies, using direct-
sequence spread spectrum. WaveLAN provides peer-to-peer communication
in 902-928 MHz (in the U. S.) and uses direct sequence spreading with the
CSMA/CA (carrier sense, multiple access, with collision avoidance) protocol.
Finally, Altair uses the Ethernet protocol (CSMA/CD, or carrier sense, multiple
access, with collision detection) and operates in microwave spectrum near 18
GHz and requires a site-specific FCC license.^
Standards such as IEEE 802.11 (which we discuss in more detail in the context
of power management features) are being developed in the U. S. and in Europe.
In general, the goal is to provide data rates exceeding 1 Mb/s and also to
support architectures with infrastructure (base stations) as well as "ad-hoc
architectures," where the terminals communicate directly with each other (peer
to peer) without the mediation of a fixed base station [PGH95].
^ F C C licenses are not required for WaveLAN and FreePort which operate in the I S M
(Industrial, Scientific and Medical) part of the spectrum. These frequency bands were origi-
nally designated for operation of equipment which generates and uses R F energy locally for
industrial, scientific and medical applications, excluding applications in the field of telecom-
munications. It WcLS later suggested to use these frequencies for local telecommunication,
such cts on-site communication.
12 CHAPTER 1
Karn [Kar91] points out deficiencies of the CSMA protocol in the context of
the wireless medium. The hidden terminal problem arises when station Y can
hear both stations X and Z but X and Z cannot hear one another. In such
a case, X and Z will collide at Y since CSMA will allow both X and Z to
transmit. The exposed terminal problem arises when station X can hear far
away station Y even though X is too far from Y to interfere with its traffic to
other nearby stations. In this case, CSMA is too conservative, it will stop X from
transmitting, wasting an opportunity to re-use channel bandwidth locally. Karn
proposes to eliminate carrier sensing, turning CSMA/CA into MACA (Multiple
Access with Collision Avoidance). If station X wants to transmit to the station
Y, it first sends a short request to send (RTS) packet and inhibits its transmitter
long enough for the addressed station to respond with a CTS (Clear to send).
Collisions are still possible for the RTS and CTS packets, but the hidden and
exposed terminal problems are avoided. Different variants of this protocol have
been implemented at Xerox and at Berkeley but there are still a number of
interesting issues, such as power control which have to be resolved.
Private Wireless Data Networks and Metropolitan

Area Wireless Networks
Several private wireless data networks provide service over wide areas using
the licensed spectrum. For example ARDIS (Advanced Radio Data Information
Service) offers service to over 400 metropolitan areas with the data rate ranging
from 8 Kb/s to 19.2 Kb/s in some areas. RAM Mobile Data off'ers service over
its MobiTex network providing coverage in 216 metropolitan areas with a data
rate of 8 Kb/s. CDPD, which we described in more detail above, offers 19.2
Kb/s maximum data rate over analog cellular. Motorola's Embarc with its
expanding set of devices (such as the Inflo receiver) provides satellite-based
information services.
Recently formed Metricom provides an innovative metropolitan-area wireless

network based on frequency-hopping spread-spectrum technology. This net-
work, called Ricochet, is a mesh-like network consisting of shoebox-sized radios,
that are mounted on existing pole-tops or buildings. It offers effective bit rates
in the range of 10-40 Kb/s over 902-928 MHz frequency band.
Infrared Technology
Infrared technology offers an alternative to the standard radio frequency com-
munication. In general, it is much cheaper to use, but it is restricted only to
very short distances and is subject to line-of-sight transmission limitations. An

invisible infrared light beam, in the frequency spectrum of laser beams (in the
terahertz range), is focused from a transmitter to the receiver over a very short
distance.
Infrared transmission is subject to the following restrictions [Bat94]:
• Transmission distance of less than 2 miles

• Line of sight limitations
• Restricted to of 16 Mb/s throughput
• Proneness to environmental disturbances, such as fog, dust, and heavy rain
The advantages of the infrared technology are as follows:
Reasonably high bandwidth

No license required
Cost effective
Capable of traversing multiple paths without interference
More secure than radio
Immune to radio frequency interference and electromagnetic interference
Satellite Networks
Mobile satellite services allow complete global coverage. In these systems, satel-
Utes play the role of mobile base stations. Satellite-based systems may be
categorized according to the orbital altitude:
• Geostationary satelites (GEOS), at an altitude of 35,786 km

• Low earth orbit satellites (LEGS), at the altitudes on the order of 1000
km
• Medium earth orbit satellites, with widely varying altitudes between
those of GEGS and LEGS.
14 CHAPTER 1
The major advantage of GEOS systems is that contiguous global coverage up to

75 degrees latitude can be provided with just 3 satellites. On the other hand,
the main drawback is a large 240-270 ms round trip propagation delay and
the higher RF power required. On the other hand, LEOS systems require less
power but require frequent handofFs.^
The bit rates for the satellite communication systems are relatively low. For
example, for Qualcomm's OmniTracks, which provides two-way communication
as well as location positioning, the downlink data rate is between 5 Kb/s and
15 Kb/s and the uplink data rate is between 55 b/s and 165 b/s.^[Kat94]
Experimental satellite systems provide higher data rates, for example NASA's
ACT satellite offers T l data rates at 1.5 Mb/s.
The Hughes Direct Broadcast Satellite System provides coverage to most of

North America. It is highly asymmetric, with a shared 12 MB/s satellite down-
link, a wire-line Internet-gateway uplink, and latencies greater than 500 ms.
Individual users may achieve up to 400 Kb/s on the downlink. The system is
wireless, but not mobile; its 24-inch satellite dish, though rapid to deploy, will
not support communication for mobile units.
Future PCS
Recently, the U. S. Federal Communication Commission (FCC) allocated band-
width in the 2 GHz range for "personal communication services" (PCS). This
bandwidth was allocated via auction, and earned the U. S. government almost
7 billion dollars. Winning bidders in markets projected to be lucrative (such
as New York and Los Angeles) bid approximately 20 dollars per person. This
amount is just for the right to use the spectrum - still higher costs must be
incurred to build the requisite infrastructure.
Two basic categories: high tier, which supports macrocells, and low tier, which
is optimized for low power and low complexity, will be supported. These two
tiers roughly correspond to the digital cellular and digital cordless categories.
At this point, the initial list of potential standards has been narrowed down
from 16 to 7 but it is still unclear as to what the standards will be and when
the initial infrastructure will be deployed.
^Note that handoffs occur here even when the terminal is not moving at all (since low-earth
orbit satellites move relative to fixed terrestrial positions).
^Note the asymmetry in uplink and downlink data rates.
One of the main challenges is how to mix continuous, real-time data such as
voice (and perhaps, in the future, video) with the alphanumeric data (such as
electronic mail, which is less time-critical). Several protocols such as PRMA
(Packet Reservation Multiple Access) [WCW91] have been proposed. These
protocols are based on the concept of reservation of slots within frames. In
PRMA, the channel bit stream is first organized into slots, such that each slot
can carry one packet from the mobile terminal to the base station. The time
slots are grouped into frames. Each slot within a frame is either available
or reserved, on the basis of a feedback packet broadcast in the previous frame
from the base station to all of the terminals. Terminals with new information to
transmit contend for access to available slots. At the end of each slot, the base
station broadcasts the feedback packets that report the results of transmission.
A terminal that succeeds in sending a packet to the base station obtains the
reservation for this slot in the subsequent frames. The base station, on failing
to receive a packet in the reserved slot, informs all the terminals that the slot
is once again available. PRMA has proved to be suitable for data and voice
communication and demonstrated efficient use of spectrum for voice.
2.3 P a l m t o p and Laptop Technology

The palmtop and laptop technology is expanding so rapidly that there is little
point in discussing specific products. We refer the reader to the web home pages
of major players on the market such as Apple, Motorola, Hewlett-Packard, NEC,
Toshiba, IBM, etc.
Instead, we concentrate on the major issues brought about by the miniaturiza-

tion of computer terminals. These are: extension of battery life, user interface,
and display issues.
Battery Life
Energy supply is the major bottleneck for mobile wireless computers. In a
recent USA Today article, longer battery life was mentioned as the feature
most desired by mobile users.^ Unfortunately, expected growth of battery life
is very slow (only 20% in the next 10 years [CB92]). Thus, energy efficiency
®The issue of energy management is nevertheless controversial. Many sceptics claim that
periodic recharging will be sufficient to make the battery limitations go away. Others point
out that, for example, the mobile terminals used in cars will use car's energy sources, not
their own batteries. We still believe that longer battery life will be an important feature
driving the market.
16 CHAPTER 1
is a necessary feature both at the level of hardware, and software. Hardware

providers are offering energy-efficient systems that switch off the background
light of the screen, power down the disk (or eliminate the disk completely in
favor of flash memory) and offer CPUs with an energy-efficient doze mode. For
example, the Hobbit chip consumes 5,000 times less energy while in doze mode
than in the active mode (250 mW in the active mode as opposed to 50 /xW in
doze mode). There is a growing pressure on software vendors to incorporate
energy management features.
To illustrate the constraint of limited available power, consider a laptop com-

puter with a CD-ROM and a display. The constant power dissipation in a
CD-ROM (for disk spinning) is approximately 1.0 W and the power dissipation
for the display is approximately 2.5 W. Assume that AA batteries are to be
used as the power source. A typical A A cell is rated to give 800 mA-Hr at 1.2
V (0.96 W-Hr). Thus, the assumed power source will last for only 2.7 hours.
To increase the longevity of the batteries, the CD-ROM and the display may
have to be powered off most of the time.
Transmitting and receiving consumes power as well. In practice, the power re-
quired for transmitting grows as a fourth power of distance between the trans-
mitter and the receiver. Powering the receiver can also drain batteries. For
example, a WaveLAN card consumes 1.7 W with the receiver on and 3.4 W
with the transmitter on. An active cellular phone consumes even more, 5.4 W
[FZ94], while consuming only 0.3 W if in standby mode.
We believe that design of energy efficient software will be one of the main
research challenges in mobile computing. Below, we summarize some of these
efforts including CPU scheduling for low power, new communication protocols,
and energy-aware application design.
Storage Technology
Flash memories constitute new, more energy efficient, storage alternatives to
disks. Flash memories consume relatively little energy and provide low la-
tency and high throughput for read accesses. However, their high cost, $30-
50/Mbyte, is still a limiting factor. Furthermore, flash memories require addi-
tional system support for erasure management since writing to flash memory re-
quires prior erasing. Additionally, the number of times the same memory block
can be rewritten is usually limited (on the order of 100,000 times). Chapter 18
provides a detailed discussion of storage alternatives for mobile computers.
User Interface and Display Issues

The two key issues in human-computer interaction are the desirability of re-
placing the keyboard with a pen-based interface and the challenge of dealing
with a small-sized display.
Pen-based interfaces have been introduced in recent years with mixed accep-
tance. The mixed results so far are due in large part to problems with hand-
writing recognition (shown by early experience with the Apple Newton Mes-
sagePad). As recognition-related technology improves, it is expected that pen-
based interfaces will gain in acceptance.
Small display size is a serious problem, especially for users who would like to
access remote information services, such as those provided on the World-Wide
Web. The need to deal with varying display sizes suggests that applications be
structured in a display independent manner. Some of these issues are discussed
in Chapter 24.
3 RESEARCH ISSUES
In this section, we review the main research issues in mobile computing and
summarize current work in the field.
As we pointed out in the previous section, the main research challenges of

mobile computing are due to mobility, variable communication conditions, and
energy limitations. We shall discuss how these challenges affect different layers
of OSI hierarchy starting from the networking layer and including some of the
discussion of the data link layer as well. Here, we make an important distinction
between the mobile communication which involve physical and MAC layers,
and mobile computing, which we broadly define as including mobile networking
(data link, network, transport layers) as well as software applications such as
wireless access to information resources, client-server interaction, etc. In this
book, we do not discuss the layers belonging to mobile communications, but
rather, restrict our attention to those belonging to mobile computing.
18 CHAPTER 1
3.1 Mobile Networking

In this section, we review how the network, transport, and data-link layer
protocols are affected in mobile and wireless environments.
Mobility Management - Network Layer

The physical location of a mobile unit no longer determines its network address.
Thus, mobility poses a major challenge from the point of view of the network
level of OSI. How does the network know where a given user is? How does the
network route messages to mobile users?
These questions are currently being addressed in two different communities:

the Internet community and the cellular-communication community. The work
in the Internet community involves mobile IP [JMP95, PB94] aiming towards a
standard which would extend IP in order to deal with mobile hosts. The work
in the cellular-communication community is the effort on location management
of cellular phone users [BI96]. The latter deals mainly with connection-oriented
communication, since it is motivated by issues of call set-up in telephony. On
the other hand, the mobile IP work assumes a connectionless, packet-switching
scenario.
The main problem of mobility management is how to find an adequate tradeoff

between searching and informing. Searching is performed by the system in case
the address of the message recipient is not known or, at least, it is not known
precisely. Informing is the activity of the mobile unit, which informs the system
about its position. There are two extreme solutions: one in which mobile units
never inform the system about their positions and another, where units always
inform the system about their movements. The second scenario works very well
for those units that receive messages frequently. In such cases, the overhead
of searching large portions of the network (possibly, even the entire network)
is avoided at the expense of the mobile unit informing the system upon each
move. If, additionally, the unit does not move between cells often, the second
strategy is a clearly a good choice. On the other hand, if the unit moves very
often and does not receive many messages, it is probably better not to inform
the system but, rather, to have the system perform a search when necessary.
There is clearly a need for a tradeoff: when mobile unit occasionally informs the
system (or specially designed home agent). The exact nature of the tradeoffs
and compromises is described in [IB92].
Since the users are mobile, there is a possibility of the user receiving the same
message twice (though in different cells) or not receiving a message at all while
in transition between two cells. Multicasting to mobile clients presents its own
challenges [AB94b]. The main issue in multicasting is how to guarantee "exactly
once" or "at least once" delivery in an efficient manner. The MCAST protocol
provided in [AB94b] offers a preliminary solution. An interesting problem in
multicasting is how to maintain efficiently a group view ~ the set of MSSs which
have, in their cells, at least one member of a given multicasting group.
Ad-hoc networking is the ultimate challenge for mobile networking. In ad-

hoc networking, mobile terminals can form networks without participation of
the fixed infrastructure. Such networks arise in rapid-deployment situations,
like emergency services at a disaster site, or military operations in a remote
area. The can also be employed in business situations such as meetings held
in venues without network infrastructure. The structure of an ad-hoc network
is highly dynamic. Routing tables may change frequently due to changing
communication conditions and power levels. A given terminal can serve as a
router between two other terminals, due to its current intermediate position at
one instant of time and no longer be in that position a short time later.
Transport Layer - Effects of Mobility and Wireless

Connection
The task of efficiently maintaining end-to-end flow and congestion control over
a mix of wired and wireless links is a difficult task due to the different charac-
teristics of the wired and wireless networks. Congestion is the main source of
packet loss on wired links, because error rates are very low. The situation is
reversed on the wireless link, where packet loss is caused mainly by high error
rates. Consequently, wireless and wired links require different techniques to
achieve reliability and flow control.
Caceres and Iftode in [CI95] (included as Chapter 7) was the first to observe
that the congestion control in TCP may cause incorrect behavior of TCP when
dealing with mobile hosts in a wireless network. The lack of acknowledgments
from moving hosts who are in the middle of handoff between cells will cause the
transmitting host to back off (slow down) exponentially, and it may take some
time for the sender to come back to the original transmission rate. This leads
to an unnecessary drop in throughput resulting from TCP's misinterpretation
of the receiver's move as network congestion. Similarly, higher error rates on
the wireless link will be interpreted by TCP as congestion, again leading to an
20 CHAPTER 1
unnecessary drop in the data rate. TCP was designed for wired networks with
low error rates and not for unreliable wireless networks with hosts which may
occasionally disconnect.
One possible solution to this problem involves extending TCP (or another trans-
port protocol) to handle wireless as well as wired links. Such a solution would
require eventual replacement of the old version of TCP by a new protocol. This
may not be practical. In order to avoid TCP modification, Badrinath et al.
[BB95a] (also discussed in Chapter 8) propose to split the TCP protocol into
two parts: one between the sender and the MSS local to the receiver and an-
other, which manages the connection between the mobile host and the MSS.
The TCP which runs between the sender and the MSS need not be modified.
Only the link between the MSS and the mobile host requires a modified proto-
col. The proposed protocol, called indirect TCP (I-TCP), has been implemented
[BB95b] and demonstrated to achieve much better throughput than the stan-
dard TCP, especially for mobile hosts which move across the non-overlapping
cells, and when the sender is located far away from the receiver.
The idea of indirect protocols can be extended further to other layers of OSI
and to include such application protocols as ftp, http (used in the World-Wide
Web), and remote procedure calls. To illustrate this further, let us consider the
http protocol, where the client has to make a separate TCP connection in order
to reach each single Web page. While this is a reasonable solution for the fixed
host, it is unacceptable for the mobile host communicating on a slow wireless
link. Indeed, since each TCP connection requires a 3-way connection set up
and a 4-way connection drop,^ the resulting overhead will be unacceptable on
a wireless link.
A better solution is to let the MSS represent a wireless client by opening and
closing such TCP connections for it. The mobile client would then open just one
http connection with its MSS. Such a solution would require writing a separate
protocol to handle the wireless link between the client and the MSS and would
leave the link between the MSS and the server unchanged. Thus, only the part
of the protocol dealing directly with the wireless and mobile host would require
modification. Making the MSS a local agent or proxy of the wireless client is
^To establish the T C P connection, the sender hcis to send a special packet with the initial
sequence number to which the receiver responds with its own sequence number. The third
packet is an acknowledgment from the sender. To close the T C P connection, four packets
have to be exchanged: the sender sends the connection closing packet,which is acknowledged
by the receiver, then the receiver also closes the connection on its side and the sender sends
an acknowledgement.
natural choice since the MSS is "close" to the wireless link and can monitor it
better than some more remote intermediary.
The concept of indirection can, in general, lead to the modification of stan-

dard client-server protocols to a client-proxy-server model, where clients are
represented (to certain extent) by proxies (their local MSS) when dealing with
remote servers. Such proxies can do most of the work for the clients and contact
them only when it is absolutely necessary. Among the issues to be resolved are
handofFs which result from mobile client moving between different cells during
a session with the server, and authorization and security issues dealing with
clients using proxies outside their home environment, etc.[BBMI93]
3.2 Information Services

As we go up in the OSI hierarchy, we have to address the issue of interactions
of mobile wireless clients with the information resources on the fixed network.
Network information resources can be accessed literally from anywhere via wire-
less links of varying quality and possibly varying tariffs.
Performance Metrics
Information content and methods of the information delivery depend on the
location of the user. Hence, user location is an important and dynamically
changing parameter. Cost of information will vary depending on the location
of the user, who may now face the choice of getting desired the information
now, but for higher cost, or later, for less cost. Finally, an additional precious
resource, battery power, has to be minimized when interacting with the fixed
network. We speculate that in the client's interactions with information services
(such as the World-Wide Web), measures such as queries per Hz and queries
per watt will be important. The analogous performance criteria in cellular
telephony is cell capacity: the number of telephone calls which may be handled
in a cell per unit of time. In the future, a query, a request for a page, will be the
analog of a telephone call in terms of performance metrics. Consequently, we
believe that in the future wireless services, the cell capacity will be measured
in terms of the number of queries which the local MSS can handle per unit of
time as well as the number of queries which a mobile client can issue per unit
of power.
22 CHAPTER 1
Issues in Dissemination of Information to Mobile

Clients
Wireless Information services will be characterized by their geographic scope.
Wide-area services such as stock market information, will be offered on a na-
tional scale. Macroservices,^^ such as weather, would be provided on a regional
basis, with regions extending to tens or hundreds of miles. The geographic
scope of micro services such as traffic conditions will extend to the size of the
current cells: a few miles in diameter. Finally, one could imagine picoservices in
an area corresponding to future picocells, perhaps hundred meters in diameter.
For example, parking availability could be provided within such a scope.
Designing support for information services for mobile users will require address-
ing the following key new issues:
• Interoperability and adaptability to networking environments ranging from

high-bandwidth wireless LANs to low-bandwidth cellular and infrared com-
munication links.
• Energy-efficient data access for terminals running on self-contained power

supplies
• Support for mobility and disconnection
• Support for active services; e.g., triggers, periodic data delivery, etc.
Below, we describe the impact of the each of the above issues on the design of
wireless information services of the future.
There are two basic modes of information dissemination to be considered: pub-

lish and provide on demand [IV94]. The latter is the traditional client-server
approach (the client submits a request and the server responds). The former
("publish") is different [IVB94b]. The server either periodically broadcasts in-
formation or sends it on a specific multicast channel [OPSS93, IV96]. Such
published data will eventually be filtered by the client. The server assists the
client in this filtering process by providing a directory of published information.
There are many advantages of the publishing mode: broadcasting the most fre-
quently accessed data items (hot spots) saves bandwidth since it cuts down the
^^The granularity of services does not directly correspond to the cell granularity (macro-
cells, microcells) and should not be confused with it.
number of separate but identical provide-on-demand requests. Also, it saves

client energy by avoiding energy-consuming uplink transmissions from client to
server. Providing the directory helps the client to tune selectively only to rele-
vant information and remain in doze mode for the rest of the time. This strategy
saves considerable amount of energy as demonstrated in [IVB94b, IVB94a]. Ex-
amples of periodically published data include stock quotes, traffic, and weather
conditions.
Each cell's bandwidth can be divided into three channels:
• uplink channel
• on-demand downlink channel
• broadcast downlink channel
How much bandwidth is allocated to each channel depends on the cell, and is
a function of the cell population and the pattern of requests. For example, one
cell may carry stock market information on its broadcasting channel, another
cell's server may decide, due to a limited interest in stocks, to provide stock
information only on demand.
In general, cells will differ in the type of radio infrastructure and consequently
by the amount of available bandwidth. Thus, not only the content, but also
structure and the mode of information presentation will be location-dependent.
There is an obvious need for a protocol suite which would make this depen-
dency hidden from the user who relocates between such cells. Some of the
services will have to be continued upon crossing cell boundaries with minimal
service disruption. Thus, there is a need to handle service handoff in a manner
analogous to the way in which cellular handoff is handled now. To ensure rapid
response to client requests, a "hot list" of items which the client accesses most
frequently will have to be re-bound to a new set of addresses when the client
enters a new cell. Therefore, each cell must provide a "directory of services,"
mapping services to their mode of their delivery (broadcast, multicast, or on-
demand) and channel. This directory must include local services as well as
global, wide-area services, and must be provided on a standard channel. The
exact mapping between services and their addresses is not visible to the user.
The system seamlessly rebinds the services according to the allocation used in
the new cell.
24 CHAPTER 1
Inadequacy of the current concept of information

services for wireless and mobile environments
There are already several efforts to modify network information services, such
as Web browsers, to deal with wireless and mobile clients. This work is cur-
rently performed by groups at MIT [K4-94], Rutgers [ABIN95], the University of
Washington [VB94], and Xerox [WSA-l-96]. Before we summarize these efforts,
we point out the main inadequacies of the current World-Wide Web protocols
for mobile clients:
• Page addressing based on the location of servers.
• Lack of presentation autonomy
• Lack of display independence and presentation adaptability
• Stateless servers
We address these issues, in turn, below.
Location-Based Addressing
In the http protocol, each page has a fixed IP address. Each client has to make
a connection (ftp connection) to that address in order to access this page. Thus,
there is only one address and one access mode (on-demand) which is currently
supported. This is inadequate for location-dependent information services for
wireless and mobile users, where the content of the presented page changes with
the user's location, and where the method of page delivery varies from cell to
cell. For example, the traffic conditions page will have different content in New
Brunswick than in Princeton, and its method of delivery may vary as well;
for instance it may be published every minute in Princeton, while provided on
demand in New Brunswick.
Thus, pages should be identified independent of the location of the server which
offers them. The best solution would be to use the same address for each
location-based page in every location, and create a binding to the local address
analogously to the way "800" telephone numbers in North America are bound
to actual telephone numbers [Che92, OPSS93, ABIN95].
Presentation Autonomy
The local MSS should have the autonomy to decide how a particular remote
service is going to be presented to users registered in its cell. Consider again
the example of the stock information service: the MSS may decide to publish
(broadcast) stock quotes for the S&PIOO stocks every minute, broadcast stock
quotes for the S&PSOO stocks every 5 minutes and provide other quotes on
demand with a cache invalidation service[BI94], which broadcasts periodically
reports only about significant changes of the stock values. MSS may also decide
that special events such as a stock reaching a new high require a special broad-
cast message which should be sent to subscribers. The remote stock server
will not and should not be aware of this autonomous decision of the MSS and
different MSSs may decide to "carry" the stock information services differently
in order to best utilize their resources and perhaps also maximize the profit.^^
Presentation Adaptability
By display independence, we mean the ability to present a page on a range of

devices including the telephone (through a voice processing card and prompted
menu interface enabling the user to enter selections from the telephone key-
pad), palmtops with small limited resolution, and powerful laptops equipped
with full color monitors. Presentation adaptability allows different presentation
of a page depending on the current load on the network, available bandwidth,
and energy resources on the client. For example each link should have resource
prerequisites associated with it and only links which are "eligible" for the cur-
rent environment should be displayed on the client's machine. In this way,
different clients may see different views of the same service. In fact, the same
client may see different views of the same service at different times in the same
cell due to the varying load on the network.
Stateless Servers
Web servers are stateless, that is, they do not keep any information related to
the state of their clients. Although this is a reasonable design for current Inter-
net users, this approach is less reasonable for low-power clients connected by
low-bandwidth links. By retaining information about a client, a server can deal
more effectively with such issues as location-dependent data and presentation
adaptability.
^^No one knows what the tariff structure will be for the future wireless services but one
may assume that the M S S will have to pay the remote server for syndicating its service and
will generate a profit by reselling this service to its local residents.
26 CHAPTER 1
Efforts to Modify Web Services for Mobile and

Wireless Clients
The DataMan group at Rutgers has developed an authoring interface [ABIN95]
to design location-aware information services. Pages are categorized as location-
dependent and location-independent. For example, the printer paige is location-
dependent and provides directions to the nearest printer from the user's current
location. Such pages are stored on multiple MSSs with possibly different con-
tents, depending on the MSS's location. Location-dependent pages are accessed
via a new mechanism, called nearcast, by associating a unique multicast address
with each such page. The mobile client accesses a location-dependent page by
multicasting a request for that page. The nearest MSS which has a copy of
the page and is a member of its multicasting group responds to the client's
request.^^ This is similar to subject-based naming in the spirit of [Che92].
The mobile client always uses the same address despite the possibility that the
underlying actual address of the requested page varies.
The scope of a page is another concept which is associated with a location-

dependent page. The scope is the set of locations in the immediate neighbor-
hood of the current location in which the page remains invariant. Thus, the
client needs to refresh the page content only when moving out of the page's
scope, rather than upon each crossing of a cell. The current system provides
support for multiple granularities of location information. In this way, the
scope of the page can be characterized, as, say "Rutgers Campus" without the
need to specify all the MSSs on campus.
The DataMan prototype provides features for autonomous presentation of a

service by an MSS. Thus, a given page can be provided either on demand or
published periodically, depending on the structure and volume of requests in
each cell.
Finally, DataMan provides page invalidation through multicasting. If a content

of a given page changes in time, a page invalidation message is multicasted by
the server on a specific, predefined multicast address.
In the Mobisaic project [VB94] (discussed in Chapter 14) at the University

of Washington, the location-dependent context of mobile hosts is abstracted
as <dynamic-variable, value> pairs. Changes to the values of such variables
^În this way, the nearcsist address plays a role analogous to "800" numbers (in North
America) which are bound to possibly different numbers depending on the area where it is
used.
are propagated to interested parties using a publisher-subscriber mechanism.

Based on this availability of "mobile computing contexts" of users, the authors
present two interesting concepts (and their implementations):
• Dynamic URL: A URL that references dynamic variables and whose

values are resolved when the URL is accessed. This allows the same URL
to specify different document contents,
• Active Documents: Documents whose contents change or become in-
valid when a specified set of dynamic variables changes.
The paper also presents the necessary extensions to HTML for authoring active
documents.
Other work at University of Washington [Wat94] (presented in Chapter 13)

involves principles for partioning functionality at the application-level between
a mobile, wireless client and a proxy on the static network. As an example of
a wireless application, the paper presents W*, a Web browser. This browser is
explicitly designed to optimize wireless communication and allow for a proxy
to cache and/or parse documents on behalf of the mobile client.
Work at MIT [K-f 94] deals with optimization of wireless link between the mobile
unit and the MSS. In order to avoid expensive page transfers, an alternative
solution is proposed in which pages are generated locally by the mobile client.
Thus, the mobile client, upon requesting a particular Web page, receives the
HTML code which it runs to locally generate the page.
3.3 Power Management

Power management is already being addressed in hardware. Features like
switching off backlighting, providing the CPU with a power-efficient doze mode,
replacing disks with more energy-efficient semiconductor memories, are just a
few examples of steps towards lowering power consumption. There is now a
growing pressure to improve energy management further at the software level.
Energy-efficient software is possible only if energy-management features already

exist on the hardware level. Thus, the CPU should have a low-energy mode, a
disk that can be spun down, a radio receiver that can be switched off when idle,
etc. Energy-aware software algorithms will take advantage of these features by
28 CHAPTER 1
maximizing the time the CPU spends in the doze mode, keeping the receiver off
most of the time, minimizing energy consuming transmissions, spinning down
the disk, etc.
There is substantial room for power saving features on the apphcation layer.
We have already discussed the publishing mode of information dissemination
in which the server periodically broadcasts "hot spots" of information without
explicit requests from the clients.
There is research underway at a variety of levels:
• Operating Systems:
- The work of [WWDS94] to schedule CPU operations for low energy

(discussed in Chapter 17)
— The work at Sun [B+93] on energy-efficient features in operating sys-
tems (discussed in Chapter 16)
• Communication Protocols:
- MAC Layer: such as in IEEE 802.11 and CDPD

— Application Layer: by using publishing mode, avoiding uplink trans-
missions, etc.
• I / O and File systems:

The work of [DCK-l-94] dealing with spinning down the disk.
The main objective of the low-energy features of 802.11 is to let the mobile
unit keep the receiver off most of the time. This is accomplished by careful
synchronization between the MSS and the mobile unit. When the MSS receives
a message addressed to one of the mobile units which reside in its cell, it does not
forward the message directly, but buffers it. Periodically, the MSS broadcasts
a "preview" which is a list of identifiers of all the mobile units which have
an outstanding message to be delivered by the MSS. The mobile unit turns
its receiver on synchronously to listen to the periodic preview. If the preview
contains the the mobile unit's identifier, then the mobile unit either keeps its
receiver on until the message is received, or it informs the MSS as to when it
wants to receive its messages. In either case, if the mobile unit receives only a
few messages a day (even the most heavy e-mail users typically do not receive
more than 50-100 messages a day) it can keep the receiver off most of the time,
saving considerable energy. For example in [IGP95], the energy savings for
the WaveLAN card are provided for a protocol which is a variation of 802.11.
With the average load of 30 messages a day, the receiver will be "on" for
approximately 10 seconds to a minute, rather than hours, leading to significant
energy savings. CDPD offers a similar low-energy feature. In [IGP95] some of
the effects of varying preview size are studied„ It would be useful to put some
of these features on the application layer, so that applications can influence the
size and the frequency with which the preview is broadcasted, based on their
tolerance of delay. It is also argued there that avoiding uplink transmissions
from the mobile unit to the MSS, which are very energy consuming (especially
in outdoor environments) is highly desirable. Therefore, solutions in which
the mobile unit stays active immediately following the preview message, which
contains unit's identifier, are preferable.
Further reduction in transmissions can be achieved by means of transport layer

modifications which cut down on the number of acknowledgements from the
mobile recipient. Perhaps, analogously to congestion control, we need power
control by controlling the number of acknowledgments depending on the power
resources of the client.
The energy management issues dealing with I/O operations are discussed in
Chapter 18, where it is demonstrated that replacing the disk with flash memory
can lead to very substantial energy savings.
3.4 S y s t e m s Issues in Mobile C o m p u t i n g

The key issue which will affect the basic design of file systems and, to a large
degree, database systems as well, is the issue of disconnection.
Operating System and File System Issues

Disconnection support is an important aspect of mobility support. Mobile users
will be often disconnected from the shared file systems and should be allowed to
work on their local file system until they reconnect again. On reconnection, the
changes to the file are propagated to the file server. The Coda file system which
is based on the Andrew file system allows for disconnected operation[KS92] (see
Chapter 19). A disconnected Coda client can continue to work by using any
data in the cache. When the client reconnects, the updates are reintegrated.
Update confiicts are handled similarly to reconnection process of a partitioned
30 CHAPTER 1
replicated database. It is possible that updates of a disconnected client will be

discarded if some other connected client has modified the files.
The Ficus replicated file system developed at UCLA[HPGP92] also supports

a form of disconnected access. In Ficus, each file is replicated to enhance
read availability. Further, it adopts a primary copy approach to replication
control. First, the update is propagated to the primary copy and later messages
are sent to the secondary copies. The exclusive-write/multiple-readers scheme
of synchronization is implemented by the use of write tokens. Only a site
with the write token can modify the file. Thus a disconnected client can be
considered as a server with write tokens for the files for which it has the primary
copy. However, on reconnection, communication to the secondary sites needs
to be accomplished. The file system proposal at Columbia [TD92] attempts
to alleviate the problem of expensive global communication of a replicated file
system by using a lazy update scheme. Further, to reduce update propagation,
two diflFerent read semantics (loose read, a best-effort value, and a strict read,
giving the most recent version) are provided.
Disconnected operation has been added to the AFS as part of the Little Work
project at the University of Michigan [HHRB92, HH93]. Before disconnec-
tion, most recently accessed data is cached at the client. Upon reconnection,
file system updates that are recorded in a log are propagated to the server.
While in disconnected mode, the mobile unit performs local operations but
network operations are logged and deferred. When the network connection
is re-established, the cache manager iterates through the log and updates are
transferred to the server. If a conflict is detected, then the replay agent notifies
the user that manual resolution is needed.
Work in this area [KS92, HPGP92, TD92] has focused on environments in which
the users are connected either over a fast network or totally disconnected from
the network. Weak connectivity (wireless connection) and energy restrictions
have not been considered.
Caching in mobile environments is addressed in [BI94]. There, the MSS does

not keep the state information about clients but rather periodically broadcasts
invalidation reports so that, upon reconnection, the mobile unit can determine
which cached items are valid and which items are invalid without performing
expensive uplink requests.
Database Systems Issues

It is clear that mobile clients will not run a complete database system lo-
cally. Most likely, mobile users will use very lightweight client software on
their machines and use remote database servers to process most queries and
transactions.
Clients will, however, perform simple queries and transactions on local data,
especially when they are disconnected. Such transactions will generally use
operations of three types:
• Operations against closely-held data, in which the chance of data conflict

is low.
• Operations that commute, such as increment and decrement.
• Operations on a partition of the actual data. An example of this is a set of
tickets where a mobile unit may be allocated a subset of the set of available
tickets prior to disconnection [SS90].
If transactions follow the above restrictions, updating the shared database at

reconnect time is straightforward. Special applications may require more com-
plex reconnection protocols and, possibly, human intervention. The impact of
mobility on transaction processing is described in [KJ96].
During periods when the mobile unit is connected, queries should be processed
in an energy-aware manner. This involves making tradeoffs between energy
used in communication and energy used locally. It may also mean choosing a
query processing plan that is suboptimal in terms of time, but optimal in terms
of energy (for example, by reducing the time the disk is spinning). This issue
is explored in detail in [AG92].
Many applications of mobile computing, especially those relating to travel

and traffic information, require the processing of location-dependent queries in
which the user's location is a parameter of the query. Processing such queries
is discussed in [IB92].
Recovery is an interesting issue since the mobile unit is subject to catastrophic

failures like theft, physical destruction, etc. Thus, there is no such thing as
truly stable storage on the mobile unit. As a result, the local log may have to
be uploaded periodically to the fixed network as described in [AB94a]. Further-
more, since much of the mobile user's interaction is of a non-database nature,
32 CHAPTER 1
it is desirable to apply database-style recoverability to a broad set of mobile

applications.
Disconnection management is even more important in the database context

than it is in the context of file systems. Because disconnection is part of routine
mobile computing, rather than a failure mode, the database buffer manager
must be able to deal with periods of disconnection without the imposition of
high-overhead at the time of reconnection. [BI94] shows new techniques for
cache invalidation suitable for mobile and wireless environments.
For many, perhaps most, applications, the user interface will be application
specific. However, general-purpose interfaces require an alternative to SQL-
based languages. Graphical user interfaces designed for workstations tend not
to map well to the small screen of mobile computers. For casual database users,
and users of PDAs, a pen-based interface is required. Preliminary work in this
area is described in [AHK92].
A more detailed discussion showing which database issues will be most affected
in mobile and wireless environments is presented in [1] as well as in [IB94].
3.5 Research Prototypes

There are several research prototypes currently being developed, many of which
are described in this book. The Ubiquitous Computing project at Xerox, the
Infopad project at Berkeley, and Coda at CMU were among the most significant
early efforts. Xerox's work is described in this book in Chapters 2, 15, and 17.
The CMU work is described in Chapter 25. The Infopad work [SLBR94] repre-
sents one extreme in the model of computation for a mobile system. It gives the
smallest role to the mobile station. While this imposes the disadvantage that
the mobile unit is useless without a network connection, it offers significant ad-
vantages in power (100 milliWatts for the entire mobile unit), and in the ability
to devote the entire device to user interaction. In this work, it is assumed that
the mobile devices ("Pads") are always connected. Computation and storage
is handled by stationary machines on the backbone (wired) network, and the
mobile device (a multimedia terminal) need handle only bitmaps sent over the
wireless network (an RF line).
4 BOOK CONTENT
In this section, we review the content of the book by briefly summarizing each
chapter o
After the first two chapters, the ordering of chapters is generally consistent with
the bottom up view of the OSI hierarchy. Thus, we begin at the network layer
by reviewing mobility management issues and ad-hoc networking (in Chapters
3 through 6), and then continue with the transport layer issues in Chapters 7
through 10. Chapters 11 through 14 discuss higher level protocols for infor-
mation services, and Chapters 15 through 21 general systems issues in mobile
computing. Finally, Chapters 22 through 25 review some research prototypes
and applications.
Chapter 2 presents a comprehensive summary of the most prominent and ex-

tensive research prototype: the Ubiquitous Computing project at Xerox. In
the Xerox model, there is no fixed attachment between the user and the mobile
terminal. Xerox's vision predicts that small computers will become as univer-
sal as paper pads and that mobile users will possibly use different terminals
at diffêrent locations. The PARCTab system integrates a palm-sized mobile
computer into an oflSce network with small-cell wireless communication. Three
different types of devices, PARCTab (a palm sized computer), PARCPad (an
electronic notepad), and Liveboard were built to illustrate three different scales
of devices. The paper presents the basic philosophy of ubiquitous computing
and describes the system design and application components of this pioneering
work. Thus, it serves as a good overall guide to the broad issues facing the
designer of a mobile system.
Mobility management is addressed in Chapters 3 and 4. Chapter 3, "Scalable

Support for Transparent Mobile Host Internetworking," by David Johnson,
describes the current status of work on Mobile IP in the form of a review of
the latest IETF proposal. This chapter reviews the main concepts introduced
by the IETF proposal such as registration, tunneling, foreign agents, etc. It
also discusses the location update mechanisms and location caching. Chapter
4, "Mobility Management in Internet and Cellular Environments," by B. R.
Badrinath and T. Imielinski, compares work on mobility management on the
Internet with the work on location management in cellular telephony. Two
distinct research communities are working on this problem. The paper attempts
to compare different approaches and points out similarities and differences in
network and system assumptions.
34 CHAPTER 1
Work on ad-hoc networking protocols is presented in Chapters 5 and 6. Chapter

5, "Dynamic Source Routing in Ad Hoc Wireless Networks," by David B. John-
son and David A. Maltz, presents a routing protocol for ad-hoc networks which
is based on dynamic source routing. An intesting feature of this protocol is that
it adapts quickly to routing changes when hosts movement is frequent, yet it
requires little overhead when hosts move less often. The overall overhead of the
protocol is measured through simulations and is determined to be quite low,
falling to just 1% of the total data packets transmitted. Chapter 6, "Routing
over Multi-Hop Wireless Network of Mobile Computers," by Charles Perkins
and Pravin Bhagawat, offers an alternative approach to routing in ad-hoc net-
works. The mobile hosts are viewed as routers which periodically advertise
their view of the interconnection topology to other mobile hosts. The key chal-
lenge is to avoid advertising unstable routes. To this end, a route updating
mechanism, which is based on the past history, is designed. The routing algo-
rithm is an extension of the today's distance vector approach (Bellman - Ford)
with additional features which reduce the impact of fluctuations in unstable
routes.
The transport-layer issues in wireless and mobile environments are discussed

in Chapters 7 through 10. Chapter 7, "Improving the Performance of reliable
Transport Protocols in Mobile Computing Environments," by Ramon Caceres
and Liviu Iftode, introduces the basic challenges of designing transport-layer
protocols for wireless and mobile environments. As we have pointed out ear-
lier, reliable transport protocols such as TCP interpret delays and losses re-
sulting from handofFs and disconnections as signs of network congestion. This
degradation is quantified in this chapter through the measurement of protocol
behavior in a wireless networking test bed. It is demonstrated that the current
TCP implementation produces unacceptably long pauses in communication dur-
ing cellular handoff's (800 milliseconds or longer). The need for differentiation
between motion-related and congestion-related packet losses is postulated and
preliminary solutions are sketched. Chapter 8, "I-TCP, Indirect TCP for Mobile
Hosts," by A. Bakre and B. R. Badrinath, is an implementation of a solution to
the problem reported in Chapter 7. I-TCP splits the end-to-end transport-layer
interaction into two parts: the fixed network part and the wireless part. The
fixed network part remains unchanged (which is beneficial since it is imprac-
tical to assume that the entire TCP will have to be changed to accommodate
mobile hosts), while the wireless part, between the MSR and the mobile host
is new. The wireless part of the I-TCP "understands" that the handoff should
not cause exponential backoff. The experimental results present significant per-
formance gains (throughput) especially for wide-area communication and with
moves occurring between nonoverlapped cells. Chapter 9 provides an extension
of TCP for wireless environment which is also based on the concept of indirec-
tion (as in Chapter 8). The proposed "Mowgli" communication architecture

is designed to split the channel with end-to-end control into two parts with a
store-and-forward type interceptor. This interceptor allows implementation of
two separate subsystems, one wireline-oriented and the other, wireless-oriented.
Thus, similarly to the work in Chapter 8, the existing TCP/IP communication
infrastructure can be retained, while the special purpose protocol has to be
implemented only for the wireless link.
Chapter 10, "Asynchronous Video: Coordinated Video Coding and Transport

for Heterogeneous Networks with Wireless Access," by J. Reason, L. C. Yun,
Allen Lao and David Messerschmidt, addresses transport-layer problems deal-
ing with the transmission of asynchronous video over wireless networks. The
main goal of this work is to minimize the perceptual delay, that is, the delay
perceived by the end user. Two related techniques are introduced in order to
reduce this perceptual delay: coding the video in such a way that there is no
need for transcoders, and asynchronous reconstruction of the video presenta-
tion at the receiver. For instance, the low motion areas are more tolerant of
delay than the high motion ones. This leads to the additional compression
through discarding fine-resolution information in the areas of high motion and
transporting information in low-motion areas less often. Analytical and simula-
tion results are presented that demonstrate substantial gains in traffic capacity.
Allowing the worst-case transport-layer delay to increase may actually increase
the overall network traffic capacity. This is particularly advantageous on wire-
less access links, though it is incompatible with the approach taken by existing
MPEG compression standards for video.
Chapters 11 through 14 discuss the higher level issues (above transport layer)
dealing with the wireless access to information. Chapter 11, "Wireless Publish-
ing: Issues and Solutions," by T. Imielinski and S. Vishwanathan introduces
the concept oi publishing into client-server interactions in wireless environments.
Publishing is a periodic broadcasting of information by the MSS over a wireless
medium. Publishing mode is intended for information which is likely to be
requested frequently and its advantages include reduction in uplink traffic, as
well as power savings on the client platform. Chapter 11 provides an overview
of different addressing methods for publishing mode, which are based either
on temporal or on multicast addressing. Addressing allows clients to tune se-
lectively to relevant pieces of published data and switch to a low-power mode
of operation (CPU in the doze mode, the receiver switched off, etc.), while no
relevant information is being published. Algorithms for adaptive scheduling of
publishing mode in conjunction with the standard "on-demand" client-server
interactions are presented as well.
36 CHAPTER 1
Chapter 12, "Broadcast Disks: Data Management for Asymmetric Communi-

cation Environments," by S. Acharya, R. Alonso, M. Franklin and S. Zdonik
views repetitive broadcast (publishing mode) as a "disk on air". The paper
describes different scheduling policies with frequency of broadcast proportional
to the access probability. Similarly to [IVB94a], periodic broadcast is viewed
as a form of a cache on air and another level of memory hierarchy.
Chapters 13 and 14 present first attempts (along with the DataMan project
described earlier) to study the design and development of location-dependent
information services.
Chapter 13, "Application Design for Wireless Computing," by Terri Watson,

presents principles for partitioning functionality at the application-level be-
tween a mobile, wireless client and a proxy on the the static network. As an
example of a wireless application, the paper presents W*, a Web client browser.
This browser is explicitly designed to optimize wireless communication and al-
low for a proxy to cache and/or parse documents on behalf of the mobile client.
Chapter 14, "Mobisaic: An Information System for a Mobile and Wireless

Computing Environment," by G. Voelker and B. Bershad, deals with primitives
necessary to build location-dependent applications. The location-dependent
context of mobile hosts is abstracted there as <dynamic variable, value> pairs.
Changes to the values of such variables are propagated to interested parties
using a publisher-subscriber mechanism. Based on this availability of "mobile
computing contexts" of users, the authors present two interesting concepts (and
their implementations): dynamic URLs and active documents. The paper also
presents the necessary extensions to HTML for authoring active documents.
Chapter 15, "Providing Location Information in a Ubiquitous Computing En-

vironment," by Mike Spreitzer and Marvin Theimer, reviews security issues
dealing with the control information about location. A new architecture is
presented in which users have primary control over location information.
Chapters 16 through 21 deal with broad systems issues of mobile computing,

ranging from the operating and file system issues to database applications for
mobile computers.
Chapter 16, "Unix for Nomads: Making Unix Support Mobile Computing,"
by Michael Bender, et al., addresses the design issues for extensions of Unix^^
(and, specifically Solaris) to support nomadic computing. It is argued there
Ûnix is a trademark of X/Open
that substantial changes ranging from the kernel level through the user com-
mand set are necessary to support mobile computing. The paper discusses the
kernel changes to support power management, the PCMCIA standard, serial
line networking and new electronic mail applications designed to specifically
deal with slow, serial line connections.
Energy-management issues are discussed in Chapters 17 and 18 (along with

Chapter 11).
Chapter 17, "Scheduling for Reduced CPU Energy," by Mark Weiser, Brent
Welch, Alan Demers, and Scott Shenker, discusses modifications in the CPU
scheduling policies to reduce energy consumption. The metric of interest is
how many instructions are executed for a given amount of energy, or MIPJ
(Instructions per Joule). Trace driven simulations are provided to compare
different classes of schedules with varying computing speeds allowing the CPU
to use lower voltage and consequently, reduce energy consumption.
Chapter 18, "Storage Alternatives for Mobile Computers," by F. Douglis, et

al., investigates the impact of new storage devices such as small hard disks,
flash-memory disk emulators, and flash memory cards on file system design.
As in Chapter 17, one of the main issues is power consumption. It is concluded
that using magnetic disks for file storage on mobile computers is basically un-
acceptable from an energy point of view. To extend battery life, it is possible
to spin down the disk when it is idle. But even then, the power consumption
is an order of magnitude higher than for a file system using flash memory. In
general, a flash-memory file system has the most attractive quality with respect
to energy and performance. However, the price and capacity limitations are
still major drawbacks.
Chapters 19 and 20 deal with the impact of disconnection on file system de-
sign. Disconnection is one of the central issues in mobile computing and Coda
was the first implementation of the file system explicitly supporting disconnec-
tion. Chapter 19, "Disconnected Operation in the Coda File System," by James
Kistler and M. Satyanarayanan, describes the design and implementation of the
Coda file system at CMU. Clients view Coda as a single, location-transparent
Unix file system. While a client is disconnected, its file system requests are
serviced solely based on the contents of local caches. File replication is used
in order to improve availability. While Chapter 19 describes design and imple-
mentation issues. Chapter 20, "Experience with Disconnected Operation in a
Mobile Computing Environment," by M. Satyanarayanan et al., presents qual-
itative and quantitative data on file access using Coda file system over a period
of about 2 years. This chapter describes the results of quantitative evalua-
38 CHAPTER 1
tions including disk space requirements, reintegration latency, and the value of
log optimization» Hoarding and hoard profiles, which describe the path names
identifying objects of interest to the user, lead to cache-content optimization.
The impact of more advanced concepts such as multi-level hoard priorities and
"meta expansions" to minimize the size of hoard profiles is described as well.
Chapters 21 and 22 deal with some of the database issues in mobile computing.
Chapter 21, "Mobility Support for Sales and Inventory Operations," by N. Kr-
ishnakumar and Ravi Jain, describes the use of an escrow mechanism to allow a
distributed server architecture in which a mobile transaction can execute using
the data held by the server to which it is directly connected. The mobile sales
and inventory applications are examples of a well-defined market segment, but
the solutions described in the paper are more general, and can apply to any
set of mobile users who share a common database. Chapter 22, "On Query
Processing Strategies for Mobile Computing," by Masahiko Tsukamoto, Rieko
Tanaka, and Shojiro Nishio, considers queries that request data from mobile
hosts or request data about mobile hosts (such as location). This work con-
siders several different query processing strategies based on either broadcast
or notification. The optimal strategy depends upon parameters of the system
(computers and network), the degree of mobility of the mobile hosts, and rate
of query submission.
Chapters 23 and 24 describe some of the ongoing research prototype work.

Chapter 23, "The Case for Wireless Overlay Networks," by Randy Katz and
Eric Brewer describes a heterogeneous architecture for future mobile informa-
tion services that is based on a wireless overlay network. In an overlay network
several possibly different wireless networks may coexist. For example, a high
bandwidth infrared network can be overlayed with a lower bandwidth radio net-
work to provide connectivity between the areas of IR coverage. The resulting
architecture is discussed in detail along with the new issues of overlay network
management and application support services. A new testbed project "Bay
Area Research Wireless Access Network (BARWAN)" is described. Chapter 24,
"The DIANA Approach to Mobile Computing," by A. Keller et al., provides an
approach to deal with heterogeneity in the mobile environment. Not only must
applications run on the wide variety of mobile platforms available (and likely
to become available), but a given instance of an application may have to move
from one platform to another - and possibly to a different kind of platform.
The DIANA system attempts to separate user interface and display manage-
ment from the underlying system and communication support. The DIANA
environment also addresses the problem of application development by allow-
ing applications to be tested in a simple environment before being interfaced
into DIANA,
Chapters 25 and 26 focus on two important applications of mobile comput-

ing: maintenance and repair systems and travel information systems. Chapter
25, "The CMU Mobile Computers as Maintenance Assistants," by Daniel P.
Siewiorek and Asim Smailagic, describes a "wearable computer" called "Vu-
Man." An earlier version of VuMan was used as a campus navigational assis-
tant. The current version is targeted at maintenance operations by the U. S.
Marines. It provides a means for the user to access maintenance manuals while
performing repairs in locations where it is not practical to access a printed
manual. (Current procedures require two people - one to do the work and
the other to read the manual.) In addition to providing manual access, this
system allows input of maintenance data via a simple user interface. Chapter
26, "Genesis and Advanced Traveler Information Systems (ATIS)," by Shashi
Shekhar and Duen-Ren Liu, pertains to the application domain of transporta-
tion systems. Unlike many of the well-publicized services that provide routing
information based on a static database of maps and real-time positional data,
the Genesis project at Minnesota is including real-time traflBc data as a param-
eter in routing decisions. This creates significant demands on the system, since
traffic data must be processed promptly and triggers must be used to inform
vehicles whose routes may be affected by accidents or reported congestion.
REFERENCES
[AB94a] A. Acharya and B. Badrinath. Checkpointing distributed applica-
tions on mobile computers. In Proc. Third International Conference on
Parallel and Distributed Information Systems, Austin, Texas^ pages 73-80,
September 1994.
[AB94b] A. Acharya and B. R. Badrinath. Delivering multicast messages in

networks with mobile hosts. In Fourteenth International Conference on
Distributed Computing Systems, June 1994.
[ABIN95] A. Acharya, B. R. Badrinath, T. Imielinski, and J. Navas. Towards
Mosaic Hke location dependent services. Technical report, Rutgers Univer-
sity, May 1995.
[AG92] R. Alonso and S. Ganguly. Energy-efficient query optimization. MITL-

TR-33-92, Matsushita Information Technology Laboratory, Princeton, NJ,
1992.
40 CHAPTER 1
[AHK92] R. Alonso, E. Haber, and H. F. Korth. A database interface for mobile

computers. In Proceedings of the 1992 Globecom Workshop on Networking
of Personal Communication Applications, December 1992.
[AK93] R. Alonso and H. F. Korth. Database system issues in nomadic com-
puting. In Proceedings of ACM-SIGMOD 1993 International Conference
on Management of Data, Washington B.C., 1993.
[B-l-93] M. Bender et al. Unix for nomads - making Unix support mobile
computing. In First Usenix Symposium on Location Dependent Computing,
August 1993. Also included in this volume.
[Bat94] Bud Bates. Wireless Networked Communications. McGraw-Hill, 1994.
[BB95a] Ajay Bakre and B. R. Badrinath. I-TCP: Indirect TCP for mobile
hosts. In 15th International Conference on Distributed Computing Sys-
tems, May 1995.
[BB95b] Ajay Bakre and B. R. Badrinath. M-RPC: A remote procedure call
service for mobile clients. In First ACM Conference on Mobile Computing
and Internetworking, May 1995.
[BBMI93] B. R. Badrinath, Ajay Bakre, Roy Marantz, and T. ImieUnski. Han-
dling mobile hosts: A case for indirect interaction. In Fourth Workshop on
Workstation Operating Systems (WWOS-IV), pages 91-97, October 1993.
[BI94] D. Barbara and T. ImieUnski. Sleepers and workoholics: caching strate-
gies in mobile environments. In Proceedings of ACM-SIGMOD 1994 In-
ternational Conference on Management of Data, Minneapolis, Minnesota,
pages 1-13, May 1994.
[BI96] B. R. Badrinath and T. ImieUnski. MobiUty management in Interenet
and cellular telephony. In this volume, 1996.
[Cal88] George Calhoun. Digital Cellular Radio. Artech House, 1988.
[CB92] Ananth Chandrasekaran and R. W. Broderson. A portable multimedia
terminal for personal communications. IEEE Communications Magazine,
pages 64-75, December 1992.
[Che92] D. Cheriton. Dissemination-oriented communication systems. Techni-
cal report, Department of Computer Science, Stanford University, 1992.
[CI95] Ramon Caceres and Liviu Iftode. Improving the performance of reliable
transport protocols in mobile computing environments. In IEEE Journal
on Selected Areas in Communications, pages 850-857, June 1995. Also
included in this volume.
[DCK-l-94] F. Douglis, Ramon Caceres, M« Frans Kaashoek, P. Krishnan, Kai

Li, Brian Marsh, and Joshua Tauber. Storage alternatives for mobile com-
puters. In SOSDI Usenix Symposium, November 1994. Also included in
this volume.
[FZ94] G, Forman and J Zahorjan. Mobile wireless computing. IEEE Spectrum,
1994.
[Goo91] David J. Goodman. Trends in cellular and cordless communications.

IEEE Communications Magazine, June 1991.
[HH93] Lo B, Huston and Peter Honeyman. Disconnected operation for AFS.
In USENIX Conference on Location Dependent Computing, August 1993.
[HHRB92] Peter Honeyman, L. Huston, J. Rees, and D. Bachmann. The Little
Work Project. In Proceedings of the Third IEEE Workshop on Workstation
Operating Systems, April 1992.
[HPGP92] J. S. Heidemann, T. W. Page, R. G. Guy, and G. J. Popek. Primarily

disconnected operation: Experiences with Ficus. In Proceedings of the
Second Workshop on the Management of Replicated Data, November 1992.
[IB92] T. Imielinski and B. R. Badrinath. Querying in highly mobile and dis-

tributed environments. In Proceedings of the Eighteenth International Con-
ference on Very Large Databases^ Vancouver, 1992.
[IB94] T. Imielinski and B. R. Badrinath. Mobile computing - solutions and
challenges in data management. Communications of ACM, October 1994.
[IGP95] T. Imielinski, M. Gupta, and S. Peyyeti. Energy efficient data com-
munication and filtering. In Proceedings of Usenix Symposium on Location
Dependent Services, April 1995.
[IV94] T. Imielinski and S. Vishwanathan. Adaptive wireless information ser-

vices. In Proceedings of SIGDB, October 1994.
[IV96] T. Imielinski and S. Vishwanathan. Wireless publishing: Issues and

solutions. In this volume, 1996.
[IVB94a] T. Imielinski, S. , and B. R. Badrinath. Energy efficient indexing on
air. In Proceedings of ACM-SIGMOD 1994 International Conference on
Management of Data, Minneapolis, Minnesota, pages 25-37, May 1994.
[IVB94b] T. Imielinski, S. Vishwanathan, and B. R. Badrinath. Power efficient
filtering of data on the air. In Proc. of EDBT, volume 779, pages 245-258,
March 1994.
42 CHAPTER 1
[JMP95] D. Johnson, A. Myles, and C. Perkins. Overview of MobilelP. In

RFC, April 1995.
[KH-94] Frans Kaashoek et al. Dynamic documents: Mobile wireless access to

WWW. In Workshop on Mobile Computing Systems and Applications,
December 1994.
[Kar91] P. Karn. MACA - a new channel access method for packet radio.
Technical report, Qualcomm, 1991.
[Kat94] Randy Katz. Adaptation and mobility in wireless information systems.

IEEE Personal Communications, pages 6-17, 1994.
[KJ96] Narayanan Krishnakumar and Ravi Jain. Mobility support for sales
and inventory applications. In this volume, 1996.
[KMM95] R. Kohno, R. Meidan, and L. Milstein. Spread spectrum access

methods for wireless communications. IEEE Communications Magazine,
33(l):58-68, January 1995.
[KS92] J. Kistler and M. Satyanarayanan. Disconnected operation in the

CODA file system. ACM Transactions on Computer Systems, 10(1),
February 1992. Also included in this volume.
[OPSS93] B. Oki, M. Pfluegl, A. Siegel, and D. Skeen. The information bus

- an architecture for extensible distributed systems. In Proc. of the 14th
ACM Symposium on Operating Systems Principles, December 1993.
[PB94] Charles Perkins and Pravin Bhagawat. Routing over multi hop wireless
network of mobile computers. In Proc. ACM SIGCOMM'94, 1994. Also
[PGH95] Jay E. Padgett, Christoph G. Gunter, and Takeshi Hatori. Overview

of wireless personal communications. IEEE Communications Magazine,
pages 28-41, January 1995.
[SLBR94] S. Seshan, M. T. Le, F. Burghard, and J. Rabaey, Software ar-

chitechture of the InfoPad System. Mobidata Workshop, November 1994.
[SS90] N.R. Soparkar and A. Silberschatz. Data-value partitioning and virtual

messages. In Proceedings of the Ninth ACM SIGACT-SIGMOD-SIGART
Symposium on Principles of Database Systems, Nashville, pages 357-367,
April 1990.
[TanSl] A. S. Tanenbaum. Computer Networks. Prentice Hall, New Jersey,

1981.
IntroducUon to Mobile Computing 43
[TD92] C. Tait and D. Duchamp. An efficient variable consistency replicated

file service. In Usenix File System Workshop, pages 111-126, May 1992.
[VB94] G. Voelker and B. Bershad. Mobisaic: An information system for mo-

bile wireless computing environment. In Workshop on Mobile Computing
Systems and Applications, December 1994. Also included in this volume.
[Wat94] Terri Watson. Application design for wireless computing. In Workshop

on Mobile Computing Systems and Applications, December 1994. Also
[WCW91] David J. Goodman W, C. Wong. A packet reservation multiple
access protocol for integrated speech and data transmission. Technical
report, Rutgers University, September 1991.
[WSA+96] Roy Want, Bill N. Schilit, Norman I. Adams, Rich Gold, Karin
Petersen, David Goldbert, John R. Ellis, and Mark Weiser. The PARCTab
ubiquitous computing experiment. In this volume, 1996.
[WWDS94] Mark Weiser, Brent Welch, Alan Demers, and Scott Shenker.
Scheduling for reduced CPU energy. In First Symposium on Operating
Systems and Design (OSDI), November 1994. Also included in this vol-
ume.

IMP Introduction-To-Mobile-Computing

Uploaded by

Copyright:

Available Formats

IMP Introduction-To-Mobile-Computing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IMP Introduction-To-Mobile-Computing

Uploaded by

Copyright:

Available Formats

1

In this chapter, we review mobile computing technology, research issues, and

2.1 What is Special about Radio-based

Fading, Noise, and Interference

It is quite normal for a mobile radio to experience fades of 40-50 dB [Cal88] in

1. Short-term multipath fading, or Rayleigh fading, which is due to

2. Long-term fading, or "radio shadows," which is caused by the topog-

The problem of short-term fading is addressed by exploiting antenna diversity.

Radio Transmission Power

1. Energy is a scarce resource in mobile units. Therefore, transmission with

2. Signals from other terminals are considered by a given terminal as inter-

The degree of attenuation also depends critically on frequency - the higher

Parameters at the Application-Level

There are other relevant parameters of the mobile computing environment to

2.2 Network Technology

Analog Cellular Systems

CDPD is currently being deployed in a number of regions in the country. It

Digital Cellular Systems

Robustness of digital networks: Resistance to noise and crosstalk. Efficient

Intelligence of the digital network

Flexibility and integration with the wired digital networks

Reduced system complexity

Higher user capacity

We distinguish between Direct Sequence Cellular CDMA (DS/CDMA) and Fre-

The Pan-European Global System for Mobile Communications

The IS-54 standard in the U. S., based on T D M A . IS-54 is a North

The IS-95 standard in the U. S, based on C D M A . The IS-95 stan-

The functionality of DECT system is closer to that of a cellular system than it

Wireless Local Area Networks

Private Wireless Data Networks and Metropolitan

Recently formed Metricom provides an innovative metropolitan-area wireless

very short distances and is subject to line-of-sight transmission limitations. An

Infrared transmission is subject to the following restrictions [Bat94]:

• Transmission distance of less than 2 miles

The advantages of the infrared technology are as follows:

Reasonably high bandwidth

• Geostationary satelites (GEOS), at an altitude of 35,786 km

The major advantage of GEOS systems is that contiguous global coverage up to

The Hughes Direct Broadcast Satellite System provides coverage to most of

2.3 P a l m t o p and Laptop Technology

Instead, we concentrate on the major issues brought about by the miniaturiza-

is a necessary feature both at the level of hardware, and software. Hardware

To illustrate the constraint of limited available power, consider a laptop com-

User Interface and Display Issues

As we pointed out in the previous section, the main research challenges of

3.1 Mobile Networking

Mobility Management - Network Layer

These questions are currently being addressed in two different communities:

The main problem of mobility management is how to find an adequate tradeoff

Ad-hoc networking is the ultimate challenge for mobile networking. In ad-

Transport Layer - Effects of Mobility and Wireless

The concept of indirection can, in general, lead to the modification of stan-

3.2 Information Services

Issues in Dissemination of Information to Mobile

• Interoperability and adaptability to networking environments ranging from

• Energy-efficient data access for terminals running on self-contained power

• Support for mobility and disconnection

There are two basic modes of information dissemination to be considered: pub-

number of separate but identical provide-on-demand requests. Also, it saves