LAN Interconnection

1
SECTION 15

• LAN Interconnection

2
Summary reading

• Repeaters, bridges, routers.
• Preventing routing loops by
• Source routing Spanning Trees
• IEEE Standards 802.x
• Interconnection level. Source routing.
• Read Chapter 13

3
LAN Interconnect

• very modest objectives
• route the frame from S to D so we
• avoid routing loops
• dont forward to another LAN un-necessarily
• minimal attempt to find a Best Path

4
LAN Interconnect

• Two approaches
• 1 Source Routing or
• 2 distributed routing
• with all routers and bridges along the path
  participating in route selection

5
Whats a Bridge??

LAN
LAN port

bridge

LAN
6
How to build a LAN bridge??1

• wire pair, or
• wire pair plus analog amplifier for gain
• copies the analog waveform
• V ft
• to the other LAN, i.e.
• incorporates knowledge of Level 1, the Physical
  Level, i.e.
• knows about some of the Physical Level

7
How to build a LAN bridge??2

• digital repeater dumb
• assumes that input waveforms represent binary 1
  or 0
• knows the encodings of 1 and 0
• searches for
• makes decisions and outputs noise-free encodings
  of 1 and 0
• understands more of the Physical Level

8
How to build a LAN bridge??3

• digital repeater - less dumb
• recognizes a frame
• checks frame syntax
• corrects some errors
• understands part of Level 2, i.e.
• frame syntax including EDAC
• media access protocol Level 2A

9
How to build a LAN bridge??4

• Vaguely Intelligent Bridge
• operates at Level 2 - Link Layer - MAC Sublayer
• hence no flow control
• does frame relay within MAC
• checks FCS
• discards bad frames
• does NOT modify S or D addresses

10
How to build a LAN bridge??5

• Pretty Smart Transparent Bridge
• Bridge as defined above, plus
• frame is copied ONLY if destination D is on the
  other side of the Bridge
• called Bridge-based not Source Routing

11
Bridge-based routing

• advantages
• greatly reduces useless traffic passed to
  destination side
• all routing done by bridges
• source need not
• be aware of topology
• participate in routing
• routing is done transparently for Source Destn
• disadvantages? problems?

12
Bridge-based routing

• more bridge complexity!, e.g.
• how to know if the destination D is on the other
  side??

13
The Situation
frame f
S D
Incoming LAN for f
B
Outgoing LAN for f
Half of a Transparent Bridge B
14
Problem

• How can B make the copy / no-copy decision? i.e.
• By watching the traffic go by, How to infer if D
  is reachable from the Outgoing Lan??
• ideas?

15
Ideas

• 1 ask the sysadmin to manually enter a table of
  destinations
• a kind of routing table
• pro easy to design
• con

16
con

• tedious, expensive, error-prone
• must be changed whenever D moves
• what if D is a wireless palmtop device?

17
Bridge forwarding table...

• 2 build the table automatically
• HOW??

18
Bridge forwarding table...

• Idea
• allow for imperfect tables as follows
• when in doubt, COPY, I.E.
• COPY unless D is known to be on the INCOMING LAN
  of this bridge
• How to get clues about location of D??

19
Where is D??

• Incoming-side or outgoing??
• Look at the Source Addresses of passing frames!

20

The Situation
frame f
S D
Incoming LAN for f
table for outgoing side
B
Outgoing LAN for f
Half of a Transparent Bridge B
21
Learning algorithm for Tables

• If f appears on the Incoming Side
• Its Source Sf must ? live on the Incoming
  Side, so
• enter Sf into the Table for the Incoming side
• delete Sf from the table for the Outgoing Side

22
And

• When in doubt, forward it
• i.e. if D is present in
• neither table or
• both tables

23
Can a transparent bridge screw up?

• there can be multiple routes from S to D

1
2
A sends f with D unknown 1 2 both copy it ... 1
and 2 both see the copies 1 2 both re-copy
it... soon, 2n copies!
A
24
What to do??

• find a single, unique path pS,D from S to D
• force all traffic S,D to follow p
• repeat for all S and for all D

25
Graph theory to the rescue!

• Spanning Tree of G rooted at R
• a tree rooted on some node R of G
• every node of G appears in the Spanning Tree
• thus provides a unique path from any node to any
  other node -- via the root R!

26
We need

• distributed algorithm to calculate a spanning
  tree
• a way to map a LAN network into a graph with
  bridges and LANs as nodes

27
Mapping the LAN...

• assume
• all LANs have unique ids
• all bridges have unique ids
• one bridge R is selected as Root-node
• Rs id is broadcast to all other bridges
• each bridge B pings R both ways to find the
  fastest direction from B to R
• its root-link
• some entity on each LAN pings R all ways to
  find the fastest direction bridge to R
• its LAN-bridge link

28

LAN1
LAN2
B1
B2
LAN3
B7
B4
B5
B3
LAN6
LAN4
B6
LAN5
29

Root selected
LAN1
LAN2
B1
B2
LAN3
B7
B4
B5
B3
LAN6
LAN4
B6
LAN5
30
broadcast B1, bridges ping B1. Ping results

1
LAN1
LAN2
B1
B2
2
1
LAN3
B7
1
1
2
2
B4
B5
B3
LAN6
2
2
2
3
This number is the ping time experienced by B6
pinging B1 on its downward port
LAN4
1
B6
2
LAN5
31
(No Transcript)
32

LANs ping the root and select Lan-bridge
links NB - DIFFERENT DELAY VALUES NOW
1
2
1
LAN1
LAN2
B1
B2
2
1
3
LAN3
B7
3
3
3
2
B4
B5
B3
LAN6
2
2
2
2
LAN4
200
B6
3
LAN5
33

• the bridge pings are different numbers from the
  lan-pings
• bridge-pings simply says how long a ping to root
  takes from this bridge in each direction
• lan-pings how long a ping to root takes from
  this lan over each link

34


Discard the black links
1
2
1
LAN1
LAN2
B1
B2
2
1
LAN3
B7
3
3
3
2
B4
B5
B3
LAN6
2
2
2
2
LAN4
200
B6
3
LAN5
35
Graph theory observation

• a tree on n vertices has n-1 edges
• a loop or circuit on n vertices has n edges
• so picking one edge per node excluding the root
  node
• must yield a tree or set of trees not a loop

36
For the graph-theoretic view

• collapse each LAN into a node
• collapse each bridge into a node
• leave the bridge and LAN links alone

37
B1
LAN1
LAN2
?
?
?
B2
?
B7
?
LAN3
?
?
?
?
B3
B4
B5
?
LAN6
LAN4
?
B6
?
LAN5
?
38
Observations

• often called the Spanning Tree Algorithm Steiner
  Tree Algorithm
• Perlman, Radia An algorithm for Distributed
  Computation of a Spanning Tree... Comp. Comm.
  Rev., Vol 15, Nr 4, Sept 1985/

39
Observations

• root selection
• manual, or
• distributed election algorithm on serial number
• a LOT of traffic transits the root!

40
Observations

• net with fast backbone
• choose links bridges on the backbone
• big net
• partition it
• spanning tree per subnet
• mark all frames table entries with tree number
• replaces ONE root with n rootlets

41
Source Routing

• let the source node do it!
• Sunshine, Carl Interconnection of Computer
  Networks, CNISDN, VOl1, Nr3, 1977

42
Good news

• simple bridges
• all of the net can be used
• not just the spanning tree
• useful for interconnected LANs ethernets, token
  rings or MANs SONET rings

43
addresses

• LAN number, station address, Level 3 address

44
Bridge routing table

• my_route_segment
• mylink_in, mybridge_id, mylink_out
• if above is a substring of routing field of a
  frame, copy the frame to mylink_out
• NO state information about the stream of frames
  in the bridge/router
• every frame must carry its entire route

45
Or..

• get rid of need to carry entire route in all
  frames?
• easy - just store a pathid
• but ...

46

• now there is state in the router
• pathid, inlink, bridge_id, outlink and...
• it is now circuit switching!
• all frames of pathid must follow the same path...

47
route discovery

• by some kind of broadcast
• all-routes bcast
• D gets a copy of f for every possible route
  S,D
• single-route bcast
• traverses every link of G exactly once
• D gets one copy of the frame
• bridge forwards a frame once per outlink or
• use a spanning tree

48
route discovery proposal

• route discovery algorithm
• S sends a frame f to D using single-route bcast.
  Minimum traffic.
• D replies with all-routes bcast
• S examines the received frames with traces
  picks one

49
source routing vs Bridging

• larger frame header route trace
• end station must be aware of bridges
• bridge computation is simple
• bridge table is simple 2 link ids
• but . . .

50
Bugs

• 28 rings in series with 3 bridges per connection
  have 328 routes
• all must be traced by route discovery
• O100 years so
• LAN nrs must be manually assigned at
  configuration time?

51
More Bugs

• Broadcast storm
• S1 bcasts f using address A1
• S2 RCVES f, cant recognize A1
• forwards f
• forwarding implies discovery implies another
  broadcast

52
More Bugs

• Meltdown
• S2 rcves a mis-addressed packet P from S1
• S2 returns an ICMP message E to S1
• E is mis-directed, triggering another bcast
• P is called a Chernobyl Packet

53
Lessons

• DOnt bcast lightly
• Dont bcast across a LAN boundary
• Dont bcast at Level 2, use Level 3
• routers can terminate the call
• dont send error messages in ill-defined
  situations
• avoid synchronous actions attacks
• 10000 workstations reset clock at midnight