Lecture 7:

Bridging & Switching

CSE 123: Computer Networks
Stefan Savage

Notice:
Project 1 is up
Last time
 How do multiple hosts share a single channel?

 Medium Access Control (MAC) protocols

 Channel partitioning (FDMA,TDMA,CDMA)
 Contention-based protocols (CSMA/CD)
Lecture 7 Overview

 Moving beyond one wire

 Link technologies have limits on physical distance
 Also frequently on number of hosts connected

 Methods to interconnect LANs

 Repeaters and bridges
 Switching
Today
 What if one wire isn’t enough?

 Interconnecting different LANs

 Hubs/Repeaters: bit-for-bit rebroadcast
 Bridges: selective rebroadcast
 Switches: multi-port selective rebroadcast
Limits of a LAN
 One shared LAN can limit us in terms of:
 Distance
» Max Ethernet segment is 2500m
 Number of nodes
» Max nodes for Ethernet is 1024
 Performance

nodes
(wire)

 What to do?
Hubs/Repeaters
 Physical layer device
 One “port” for each LAN
 Repeat received bits on one port out all other ports

LAN1 Hub LAN2

LAN3
Hub Advantages
 Hubs can be arranged into hierarchies
 Ethernet: up to four hubs between any pair of nodes

 Most of LAN continues to operate if “leaf” hub dies

 Simple, cheap

7
Still One Big LAN (i.e., a Bus)
 Single collision domain
 All hosts compete for access to same logical medium
 No improvement in max throughput
 Average throughput < as # of nodes increases
 Why?

 Still limited in distance and number of hosts

 Collision detection requirements
 Synchronization requirements

 Requires performance homogeneity

 Can’t connect 10 Mbps and 100 Mbps networks

8
Bridges to the rescue
 Store and forward device
 Data-link layer device
 Buffers entire packet and then rebroadcasts it on other ports

 Creates separate collision domains

 Uses CSMA/CD for access to each LAN (acts like a host)
 Can accommodate different speed interfaces (issues?)
 Improves throughput (why?)

 Can significantly improve performance

 Not all frames go everywhere. (Why did they with a hub?)

9
Key value:
Selective Forwarding
 Only rebroadcast a frame to the LAN where its
destination resides
 If A sends packet to X, then bridge must forward frame
 If A sends packet to B, then bridge shouldn’t

LAN 1 LAN 2

A W

B X
bridge
C Y

D Z

10
Forwarding Tables
 Need to know “destination” of frame
 Destination address in frame header (48bit in Ethernet)
 Need know which destinations are on which LANs
 One approach: statically configured by hand
» Table, mapping address to output port (i.e. LAN)
 But we’d prefer something automatic and dynamic…

 Simple algorithm:
Receive frame f on port q
Lookup f.dest for output port
If f.dest found
then if output port is q then drop /* already delivered */
else forward f on output port;
else flood f;    
/* forward on all ports but the one where frame arrived*/

11
Learning Bridges
 Eliminate manual configuration by learning which
addresses are on which LANs Host Port
A 1
 Basic approach B 1
 If a frame arrives on a port, then associate its source C 1
address with that port
D 1
 As each host transmits, the table becomes accurate
W 2

 What if a node moves? X 2

Y 3
 Table aging
Z 2
 Associate a timestamp with each table entry
 Refresh timestamp for each new packet with same
source
 If entry gets too stale, remove it
12
Learning Example
Suppose C sends frame to D and D replies back with frame to C

 C sends frame, bridge has no info about D, so floods to both LANs

 bridge notes that C is on port 1
 frame ignored on upper LAN
 frame received by D

13
Learning Example

 D generates reply to C, sends

 bridge sees frame from D
 bridge notes that D is on port 2
 bridge knows C on port 1, so selectively forwards frame via port 1
14
Network Topology
 Linear organization
 Inter-bridge hubs (e.g. CS)
are single points of failure
 Unnecessary transit
(e.g. EE<->SE must
traverse CS)

 Backbone/tree
 Can survive LAN failure
 Manages all inter-LAN
communication
 Requires more ports

15
An Issue: Cycles
A
 Learning works well in B
tree topologies B3

C B5

 But trees are fragile B2

D B7

 Net admins like E F K

redundant/backup paths
B1

 How to handle Cycles? G H

 Where should B1
forward packets destined B6
B4
for LAN A?

16
Potential solutions
 Don’t allow redundant links (no loops allowed)
 E.g., physical tree… downside?

 Distributed routing protocol (SPF)

[future lecture]

 Create a temporary “virtual tree” on the physical

topology
 Spanning Tree algorithm
Spanning Tree
A
 Spanning tree uses B
subset of bridges so B3

there are no cycles C B5

 Prune some ports D B7

B2 K
 Only one tree
E F

B1
 Q: How do we find a
G H
spanning tree?
 Automatically! B6
B4
 Elect root, find paths I
J

18
Spanning Tree Algorithm:
10,000 foot view
 Elect a root node of the tree (lowest address)

 Grow tree as shortest distances from the root (use

lowest address to break distance ties)
 All bridges send periodic configuration messages over ports
for which they are the “best” path
 Then turn off ports that aren’t on the “best” paths
Spanning Tree Algorithm:
Details
 Each bridge sends periodic configuration messages
 (RootID, Distance to Root, BridgeID)
» “I am BridgeID, the Root is named RootID, I’m X hops away”
 All nodes think they are the root initially

 Each bridge updates route/Root upon receipt

 Smaller root address is better, then shorter distance
 To break ties, bridge with smaller address is better
 Record best config heard via each port

 Rebroadcast new config only to ports we’re “best”

 Don’t bother sending config to LANs with better options
 Add 1 to distance, send new configs where still “best”
 Only forward to ports to route or where we’re best

20
Spanning Tree Example
 Sample messages to and from B3: A
B
B3
1. B3 sends (B3, 0, B3) to B2 and B5
C B5
2. B3 receives (B2, 0, B2) and (B5, 0,
B5) and accepts B2 as root D B7
B2 K
3. B3 sends (B2, 1, B3) to B5
4. B3 receives (B1, 1, B2) and (B1, 1, E F
B5) and accepts B1 as root
5. B3 wants to send (B1, 2, B3 ) but B1
doesn’t as its nowhere “best”
6. B3 receives (B1, 1, B2) and (B1, 1, G H
B5) again and again…
B6 B4
Data forwarding is turned off for LAN A I
J

21
Important Details
 What if root bridge fails?
 Age configuration info
» If not refreshed for MaxAge seconds then delete root and
recalculate spanning tree
» If config message is received with more recent age, then
recalculate spanning tree
 Applies to all bridges (not just root)
 Temporary loops
 When topology changes, takes a bit for new configuration
messages to spread through the system
 Don’t start forwarding packets immediately -> wait some time
for convergence
 Broadcast packets?
 Sent on every active port
22
Switched Ethernet
 Hosts directly connected to a bridge
 learning + spanning tree protocol

 Bridge supports parallel forwarding

 A-to-B and A’-to-B’ simultaneously
 Generally full duplex as well
(A->B and B->A in parallel)

 Switch backplane capacity varies

 Ideally, nonblocking
 I.e., can run at full line rate on all ports

 No longer any shared bus

 Each link is its own collision domain
 Collision detection largely irrelevant

23
Some switching details
 Cut through switching optimization
 Only buffer packet header (for output port lookup)
» then forward remaining bits directly
 Reduced latency, but may forward bad packets
 Extremely common in real Ethernet switches

 Backpressure flow control

 Input port=1Gbps, output port = 100Mbps
 Buffer can only absorb temporary bursts
 Send JAM signal on input port when buffer gets too full
Modern extension: VLANs
 Scaling problem with switches
 All LANs in same broadcast domain
» Recall: broadcast packets sent everywhere
 As # hosts grows, broadcast traffic becomes an issue
 Virtual LANs (VLANs) created to address this issue
 Each port optionally configured with a VLAN ID
» Configured statically
 Inbound packets “tagged” with this ID
 All switches will only forward on ports that are part of the
same VLAN
 Create independent broadcast domains within a single
tree
Summary:
Layer 2 Forwarding
 Create spanning tree across LANs
 Learn which ports to use to reach which addresses

 Benefits
 Higher link bandwidth (point-to-point links)
 Higher aggregate throughput (parallel communication)
 Improved fault tolerance (redundant paths)

 Limitations
 Requires homogeneous link layer (e.g. all Ethernet)
 Can’t control forwarding topology

 What if we want to connect different link layers?

26
For Next Time

 Read 3.2 in P&D

 Reminder: Project #1 has been assigned

27