Computer Networks

P2P Overlay Networks

Mayank Pandey
Overlay Networks
• An overlay network is
a computer network
– that is built on top of
another network
• Nodes in the overlay
network can
– connected by virtual or
logical links
• Through many physical
links, in the underlying
network Computer Networks: A Systems Approach, Peterson and Davie

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 2

Overlay Networks
• Logical network implemented on top of some
underlying network:
– the Internet was built as an overlay over telephone
network
• The links that connect the overlay nodes
– implemented as tunnels through the underlying network
– For example, Tunneling in:
• VPN (Virtual Private Networks)
• IPv6 deployment (IP in IP tunnel)
• Multicast Back Bone (MBONE, IP in IP tunnel)

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 3

 Overlay Networks

Computer Networks: A Systems Approach, Peterson and Davie

• Three overlay nodes (A, B, and C) connected by a pair of tunnels.

• Overlay node B makes a forwarding decision for packets from A
to C based on the inner header (IHdr)
– and then attaches an outer header (OHdr) that identifies C as the
destination in the underlying network.

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 4

Overlay Networks

Computer Networks: A Systems Approach, Peterson and Davie

• Nodes A, B, and C are able to interpret both the inner

and outer header
– intermediate routers understand only the outer header.
• A, B, and C have addresses in both the overlay
network and the underlying network
– not necessarily the same
10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 5
Overlay Networks

Computer Networks: A Systems Approach, Peterson and Davie

• Underlying address (of A, B,C) might be a 32-bit IP address

• Their overlay address might be an experimental 128-bit
address.
– In fact, the overlay need not use conventional addresses at all
– may route based on URLs, domain names, an XML query, or even the
content of the packet.

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 6

Application Layer Overlay Network

• Also known as P2P overlays

– Nodes-> End hosts (Processes running there)
• May act as routers of network
– Edges-> Transport Layer Virtual links

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 7

Applications:
Application Service Description Examples
File storage and sharing: Gnutella,
Bittorrent, InterPlanetary File System (IPFS)
Scalable approaches to Content delivery networks (CDNs): Akamai,
Content sharing data with users Limelight
Distribution across the Internet Streaming media: Spotify, Sonos
Software update distribution: Linux, World
of Warcraft
Privacy and censorship resistance: Tor,
Distributed Delegating the work for an Freenet
Computing application across many Cryptocurrency: Bitcoin
computers Botnets and malware: Storm
Collaboration Providing real-time human Voice Over IP (VOIP): Skype
communication Instant Messaging: Tox
Platforms Building applications Java: JXTA

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 8

 P2P Networks: Overview
• Music-sharing apps:
– Napster and later on KaZaA
• introduced term “peer-to-peer”
– In context of sharing MP3 files
• Don’t download music from a
central site:
– access music files directly from whoever
on Internet have a copy on their
computer

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 9

P2P Networks: Overview
• Generally peer-to-peer network allows:
– community of users to pool and their resources
• content, storage, network bandwidth, CPU etc.
– in a distributed and decentralized manner

• P2P networks provide access to larger

– archival store, computation capability , Audio/video
streaming and conferencing
• than any one user could afford individually.

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 10

P2P Networks: Overview
• Decentralized and Self-Organized:
– nodes organize themselves into a network
• without any centralized coordination
• Locating an object of interest and downloading that
– happen without centralized authority
• System is able to scale to millions of nodes
– Nodes are hosts willing to share objects
– Transport layer links connecting these nodes
• Utilized to traverse sequence of machines to locate object

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 11

P2P Networks: Types
• Two main P2P Network types
– Unstructured (based on searching)
“Network has structure, but peers are free to join any where
and objects can be stored anywhere”
– Structured (based on addressing)
“Network structure determines where peers belong in the
network and where objects are stored “

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 12

P2P Networks: Another Classification
• 1st generation
– Typically: Napster
• Only file download P2P, Publish and Search Centralized
• 2nd generation
– Typically: Gnutella
• Pure P2P
• 3rd generation
– Typically: Super-peer networks
• Hybrid of P2P and Centralized
• 4th generation
– Typically: Distributed hash tables
• Structured P2P overlays

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 13

P2P Networks: Beginning
• A killer application: Napster
– Free music over the Internet
• Key idea:
– share the content, storage and bandwidth of
individual (home) users
• Each user stores a subset of files
• Each user has access (can download) files from
all users in the system

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 14

Napster: Challenges
• Find where a particular file is stored
– Scalability: up to hundred of thousands or millions
of systems
– Dynamicity: systems can come and go any time

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 15

Napster: History
• The rise
– January 1999: Napster version 1.0
– May 1999: company founded
– December 1999: first lawsuits
– 2000: 80 million users
• The fall
– Mid 2001: out of business due to lawsuits

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 16

Napster Technology
• User installs the software
– Download the client program
Bob
– Register name, password, local directory, etc. centralized
• Client contacts Napster (via TCP) directory server
1
– Provides a list of music files it will share peers
1
– Napster’s central server updates the directory
• Client searches on a title or performer 1 3

– Napster identifies online clients with the file 2 1

– and provides IP addresses

• Client requests the file from the chosen
supplier Alice
– Supplier transmits the file to the client
– Both client and supplier report status to Napster

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 17

Napster: Properties
• Publishing and Searching Centralized
– Only download in P2P manner
• No load on single provider
• Server’s directory continually updated
– Always know what music is currently available
– Point of vulnerability for legal action
• Proprietary protocol
• Bandwidth issues
– Suppliers ranked by apparent bandwidth & response time
• Single point of failure, Performance bottleneck, Copyright
infringement

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 18

Pure P2P system: Gnutella
• Gnutella history • Query flooding
– 2000: J. Frankel & – Join: contact a few
T. Pepper released nodes to become
Gnutella neighbours
– Soon after: many other • Publish: no need!
clients (e.g., Morpheus,
• Search:
Limewire, Bearshare)
– ask neighbours, who ask
– 2001: protocol
their neighbours
enhancements, e.g.,
“ultrapeers” • Fetch: get file directly
from another node
10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 19
Gnutella: Query Flooding
• Fully distributed • Overlay network:
– No central server graph
• Public domain protocol – Edge between peer X
– Many Gnutella clients and Y if there’s a TCP
implementing protocol connection
– All active peers and
edges form the overlay
network
– Given peer will typically
be connected with < 10
overlay neighbors

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 20

Gnutella: Protocol
• Query message sent
File transfer:
over existing TCP HTTP
connections
• Peers forward Query
QueryHit
Query message
ery Qu
it
• QueryHit sent over Qu ryH ery
ue
reverse path Q
Query
QueryHit

Scalability: Qu
er
y
limited scope
flooding

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 21

Gnutella: Peer Joining
• Joining peer X must find some other peers
– Start with a list of candidate peers
• A bootstrap server is needed
– X sequentially attempts TCP connections with peers on
list until connection setup with Y
• X sends Ping message to Y
– Y forwards Ping message.
– All peers receiving Ping message respond with Pong
message
• X receives many Pong messages
– X can then set up additional TCP connections

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 22

Gnutella: Pros and Cons
• Advantages
– Fully decentralized
– Search cost distributed
– Processing per node permits powerful search
semantics
• Disadvantages
– Search scope may be quite large
– Search time may be quite long
– High overhead, and nodes come and go often

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 23

 Hybrid P2P system: KaAzA
• KaZaA history
– 2001: created by Dutch company (Kazaa BV)
• Smart query flooding
– Join: on start, the client contacts a super-node (and
may later become one)
– Publish: client sends list of files to its super-node
– Search: send query to super-node, and the super-nodes
flood queries among themselves
– Fetch: get file directly from peer(s)

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 24

KaZaA: Exploiting Heterogeneity
• Each peer is either a group
leader or assigned to a
group leader
– TCP connection between
peer and its group leader
– TCP connections between
some pairs of group leaders
• Group leader tracks the
content in all its children
ordinary peer

group-leader peer

neighoring relationships
in overlay network
KaZaA: Motivation for Super-Nodes
• Query consolidation
– Many connected nodes may have only a few files
– Propagating query to a sub-node may take more time than
for the super-node to answer itself
• Stability
– Super-node selection favors nodes with high up-time
– How long you’ve been on is a good predictor of how long
you’ll be around in the future

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 26

 BitTorrent
• BitTorrent history and motivation
– 2002: B. Cohen debuted BitTorrent
– Key motivation: popular content
• Popularity exhibits temporal locality
– Focused on efficient fetching, not searching
• To handle very large files
• If few publisher and many downloaders
– Publishers may bog down

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 27

BitTorrent: Simultaneous Downloading
• Divide large file into many pieces
– Replicate different pieces on different peers
• Peer can (hopefully) assemble the entire file
• Allows simultaneous downloading
– Retrieving different parts of the file from
different peers at the same time
• And uploading parts of the file to peers
– Important for very large files

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 28

BitTorrent: Simultaneous Downloading
• Replication: natural side effect of downloading
– As soon as a peer downloads a particular piece
• it becomes another source for that piece.
– The more peers downloading pieces of the file, the
more piece replication occurs
• distributes the load proportionately
• more total bandwidth is available to share the file with
others.
• Pieces are downloaded in random order
– to avoid a situation where peers find themselves lacking the same set
of pieces.

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 29

BitTorrent: Swarms
• File is shared via independent BitTorrent network swarm
• The lifecycle of a swarm
– starts as a singleton peer with a complete copy of the file
– node that wants to download joins the swarm
• becomes its second member and begins downloading pieces
– becomes another source for the pieces it has downloaded
• Other nodes join the swarm and begin downloading pieces from
multiple peers, not just the original peer
– If the file remains in high demand
• New peers keep on replacing those who leave the swarm,
• The swarm could remain active or it could shrink back original peer
based on the demand and popularity

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 30

Swarm:

Computer Networks: A Systems Approach, Peterson and Davie

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 31

 BitTorrent Components
• Seed: Peer with entire file
• Fragmented in pieces
• Leacher: Peer with an incomplete copy of the file
• Torrent file: Passive component
• Contains meta information about file and swarm
– The target file’s size
– The piece size
– SHA-1 hash values pre-computed from each piece
– The URL of the swarm’s tracker
• Tracker: server that tracks a swarm’s current
membership
• Allows peers to find each other
• Returns a list of random peers
10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 32
32
Structured P2P: The DHTs
• Why we need DHTs?
– Searching in P2P networks is not efficient
• Either centralized system with all its problems
• Or distributed system with all its problems
– Actual file transfer process in P2P network is scalable
• File transfers directly between peers
– Searching does not scale in same way
• Original motivation for DHTs
– More efficient searching and object location
• Put another way:
– Use addressing instead of searching

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 33

Recall: Hash Tables
• Hash tables are a well-known data structure
– Hash tables allow insertions, deletions, and finds in constant
(average) time
• Hash table is a fixed-size array
– Elements of array also called hash buckets
• Hash function maps keys to elements in the array
• Properties of good hash functions:
– Fast to compute
– Good distribution of keys into hash table
– Example: SHA-1 algorithm

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 34

Hash Tables: Example

• Hash function: hash(x) = x mod 10

• Insert numbers 0, 1, 4, 9,16, and 25
• Easy to find if a given key
• is present in the table

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 35

DHT: Idea
• Hash tables are for fast
lookups
– Idea: Distribute hash buckets
to peers
• Result is Distributed Hash
Table (DHT)
– Need efficient mechanism for
• finding which peer is responsible
for which bucket and
• routing between them

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 36

DHT: Principle
• In a DHT, each node is
responsible for
– one or more hash buckets
• As nodes join and leave
– responsibilities change
• Nodes communicate to
– find the responsible node

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 37

DHT: Principle

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 38

DHT: Examples
• Chord
• CAN
• Tapestry
• Several others exist too
• Pastry, Plaxton, Kademlia, Koorde, Symphony,
P-Grid, CARP, …

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 39

Chord
• Chord was developed at MIT
– Originally published in 2001 at SIGCOMM
conference
• Paper has mathematical proofs of correctness and
performance
• Many projects at MIT around Chord
– CFS storage system
– Ivy storage system
– Plus many others…

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 40

Chord: Basics
• Chord uses SHA-1 hash function
– Results in a 160-bit object/node identifier
• Same hash function for objects and nodes
– Node ID hashed from IP address
– Object ID hashed from object name
• Object names: Based on the context
• SHA-1 gives a 160-bit identifier space
– Organized in a ring which wraps around
– Nodes keep track of predecessor and successor
• Node responsible for objects between
– its predecessor and itself
• Overlay is often called “Chord ring”
10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 41
Chord: Different procedures
• Joining the network
• Storing in to the network
• Retrieving from the network
• Periodic stabilization
– To keep the overlay structure intact

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 42

Joining: Step by Step Example
• Existing network with
nodes on 0, 1 and 4
– Hash id of nodes are 0,1
and 4
• Many different ways to
implement Chord
– Here only conceptual
example
– Covers all important
aspects
10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 43
Joining: Step by Step Example
• New node wants to join
– Hash of the new node: 6
• Known node in network:
– Node1
• Contact Node1
– With own hash id

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 44

Joining: Situation Before Join

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 45

 Joining: Contact known node

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 46

Joining: Join gets routed along the network

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 47

Joining: Successor of New Node Found

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 48

Joining Successful + Data Transfer
• Joining is successful
– Old responsible node
transfers data that
should be in new node
– New node informs
Node4 about new
successor (not shown)

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 49

 Joining: All Is Done

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 50

Storing a Value
• Node 6 wants to
store object with
name “Foo” and
value 5
• hash(Foo) = 2

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 51

 Storing a value: (Contd.)

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 52

 Storing a value: (Contd.)

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 53

Retrieving a Value
• Node 1 wants to get
object with name “Foo”
– hash(Foo) = 2
– Foo is stored on node 4

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 54

Handling Dynamism: Joining
• Stabilization Algorithm:
– Ask successor to tell about
predecessor

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 55

Handling Dynamism: Joining
• Stabilization Algorithm:
– Ask successor to tell about
predecessor
• Suppose 13 joins and sets
its successor as 15 and
predecessor as nil

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 56

Handling Dynamism: Joining
• Stabilization Algorithm:
– Ask successor to tell about
predecessor
• Suppose 13 joins and sets
its successor as 15 and
predecessor as nil
• When 13 runs stabilization
– 15 Sets its pred as 13

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 57

Handling Dynamism: Joining
• Stabilization Algorithm:
– Ask successor to tell about
predecessor
• Suppose 13 joins and sets its
successor as 15 and
predecessor as nil
• 13 runs stabilization
– 15 Sets its pred as 13
• 11 runs stabilization
– Sets its succ as 13

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 58

Handling Dynamism: Joining
• Stabilization Algorithm:
– Ask successor to tell about
predecessor
• Suppose 13 joins and sets its
successor as 15 and predecessor as
nil
• 13 runs stabilization
– 15 Sets its pred as 13
• 11 runs stabilization
– Sets its succ as 13
– 13 sets its pred as 11

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 59

Handling Dynamism: Leaving
• Some more state needed
– Successor list
• Containing some immediate K successors
• Some algorithm needed
– Check predecessor (heartbeat messages)
• If fails---- set as nil (stabilization can handle)
– During stabilization if successor does not respond:
• Replace with closest alive successor

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 60

Chord: Scalable Routing
• Routing happens by passing message to successor
• What happens when there are 1 million nodes?
– On average, need to route 1/2-way across the ring
– In other words, 0.5 million hops! Complexity O(n)
• How to make routing scalable?
– Answer: Finger tables
• Basic Chord keeps track of predecessor and successor
• Finger tables keep track of more nodes
– Allow for faster routing by jumping long way across the ring
– Routing scales well, but need more state information
• Finger tables not needed for correctness, only for performance
improvement

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 61

Chord: Finger Tables
• In m-bit identifier space, node has up to m fingers
– Fingers are stored in the finger table
• Row i in finger table at node n contains first node s
– that succeeds n by at least 2i-1 on the ring
• In other words:
– finger[i] = successor(n + 2i-1)
• First finger is the successor
– Distance to finger[i] is at least 2i-1

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 62

Chord: Scalable Routing
• Finger intervals increase
with distance from node
n
– If close, short hops and if
far, long hops
• Two key properties:
– Each node only stores
information about a
small number of nodes
• Example has five nodes
at 0, 2, 5, 6 and 11
– 4-bit ID space --> 4 rows
of fingers

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 63

Chord: Final Words
• Search performance of “pure” Chord is O(n)
– Number of nodes is n
• With finger tables
– need O(log n) hops to find the correct node

10/24/2024 Mayank Pandey, MNNIT, Allahabad, India 64