Title: WAN Technologies and Routing Realizing Physical Networks over Wider Distance
1WAN Technologies and Routing Realizing Physical
Networks over Wider Distance
2Why WANs?
- While you can extend LANs with bridging/repeaters,
paradigm starts to break down for many endpoints
and large traffic - WANs must provide
- Open ended growth - can keep adding endpoints
- Reasonable performance for large size networks
- WANs built with packet switched networks glued
together
3Packet Switches
- What is a packet switch? use a local star
topology around a single computer connects
locally and passes on to other PSs - Move complete packets from one connection to
another - Does more than replicate bits
- Each switch really a special purpose computer
- Has CPU, memory, NICs
- Now we can connect Packet Switches (over
distance) and create networks
4WANs
- Need not be symmetric
- Links can have different data rates
- Can have more than one link between switches for
capacity or redundancy
5WAN Picture
6Store and Forward
- Packets buffered in memory when received
- Placed in switchs memory (stored)
- CPU notified, interrupted
- CPU examines and determines which interface to
send packet to (forwarded) - If output interface busy packet queued
7Addressing
- WANs use hierarchical routing to make forwarding
more efficient - Addresses broken into parts
- Only fraction of address used to determine
outgoing interface - Typical two parts, first part chooses interface
ltPacket Switch number, Computer numbergt
8Addressing Picture
9Next Hop
- If destination local, send directly to given
local computer - If destination not local must forward to another
switch need get the route - Total route not kept in each switch
- Instead information on the next switch in line is
kept - Similar to airline flight lists
10Next Hop Picture
11Source Independence
- Next hop does not depend on source or path so far
- Only on destination
- Routing at airports similar
- People from multiple sources headed for multiple
destinations only care about next flight
12Source Independence Continued
- Only one table required
- Only destination information need be extracted
- All forwarding handled uniformly
- For forwarding, get the packet switch number out
of address, then index into table
13Routing
- Next hop information contained in routing table
- Process of moving packet to next hop called
routing - All packets with same first part of destination
sent to same next hop - Table indexed by destination address
- No searching, quicker just indexing
14Routing Continued
- Table contains list of destination packet
switches, not computers - Smaller amount of data to keep up
15Route Table
16Routes in a WAN
- WAN modeled as a graph
- Nodes are routers (the packet switches in
network) - Links (edges) are connections
- Edges can have weights
17WAN Graph
18Routing Table
19Default Route
- A lot of redundant information
- Router may only have single connection
- Or vast majority of traffic destined for single
router - This is a common situation
- Routing traverses the table in order so we can
have a default at the bottom
20Default Route Continued
- Have default route for most traffic
- Can have only one default route
- Has lowest priority
- All non-default checked first
- If nothing matches use default
21Default Routes
22Table Computation
- Two approaches
- Static entered once and never changes
- Dynamic continually computed
- Most use dynamic
- Monitor network connections and respond
23Shortest Path
- If have graph with weights can use Dijkstras
algorithm to compute shortest path between two
nodes algorithm flexible for weight usage - Can use result to compute next hop
- Shortest path can be defined various ways
depending upon weights we use distance is sum
of weights along a path - Can use weights to represent various costs
capacity, distance, etc. - Can use this algorithm uniformly across whole
network once and for all
24Distributed
- Each router can create table and send to
neighboring routers - One of best known is distance-vector think of
the sum of the weights of all edges of a path
(route) as the distance of that route - Each router periodically sends route information
to immediate neighbors - Each message contains destination and distance
- If I find an immediate neighbor that has a
shorter route to a destination than I have, I
change my route table data to use that! (Actually
a little more detailed the sum of my distance
to that neighbor plus his distance to ultimate
destination has to be minimal)
25Graph with Weighted Edges
A graph with weights assigned to edges. The
shortest path between nodes 4 and 5 is shown
darkened. The distance along the path is 19, the
sum of the weights on the edges.
26Computing Distance Vector Table and Abbreviated
Routing Table(For the network of the previous
slide from the perspective of Packet Switch 6)
27Distributed Continued
- Each neighbor updates own table if shorter route
exists - Alternative link-state
- AKA shortest path first
- Instead of destination and distance, status of
link between switches sent - Each switch then builds graph using Dijkstras
algorithm
28Examples
- ARPANET
- 56Kbps leased lines
- Basis for Internet
- Started with distance/vector now uses link-state
- X.25
- ITU/CCITT character oriented
- Mostly in Europe
- In US was mostly terminal to host
29Example Continued
- Frame relay
- Block up to 8K octets
- Designed to bridge LANs
- 1.5 Mbps or 56Kbps
- SMDS
- Switched Multi-megabit Data Service
- Faster than FR, small header big payload (9188
octets), very fast - Need special interface hardware
- ATM
- Asynchronous Transfer Mode
30Exercise
From the perspective of Packet Switch C in the
graph below, compute and show the distance vector
table and the abbreviated routing table that C
must use. Assume the weights given on each edge
of the graph.
1
B
F
3
1
G
A
1
3
E
C
3
1
1
D