Spring 2006

1
Spring 2006 EE 5304/EETS 7304 Internet Protocols
Lecture 4 Bridges
Tom Oh Dept of Electrical Engineering taehwan_at_engr
.smu.edu
2
Administrative Issues

• I have posted the download instruction for Cygwin
  and Putty today. Cygwins download instruction
  is outdated but I think that you can follow the
  instruction.
• I will try to list the page numbers from our text
  book relevant to each topic
• I will post the second home this week.

3
Outline

• Bridges (Pg. 164 Comer)
• Transparent learning bridges (Pg. 165 Comer)

4
Bridges to Interconnect LANs

• Work at MAC sublayer (data link layer)
• Assuming higher layers are same
• Motivation practically, each LAN may be limited
  in number of stations, size, and total bandwidth

5
Advantages of Bridges

• Overcome physical constraints users on distant
  LANs can communicate, total number of stations
  can increase by adding LAN segments
• Allows different user communities to have
  separate LANs but stay interconnected
• In this case, intra-LAN traffic is much more than
  inter-LAN
• Bridges can monitor and manage each LAN segment

6
Advantages (cont)

• More reliable and serviceable
• Fault in a LAN segment can be isolated from other
  segments
• More security is possible
• Main disadvantage is performance
• Frames going through bridges may be delayed,
  lost, misordered, or errored

7
Transparent Learning Bridges (IEEE 802.1)

• Function is to make interconnected LANs look like
  one LAN to all stations, by store-and-forwarding
• Bridge reads all frames on LAN A and stores those
  addressed to LAN B
• Waits for free channel, retransmits buffered
  frames onto B using its MAC protocol
• Does same thing from LAN B to LAN A

8
Transparent Learning Bridges (cont)

• Advantages
• No extra work or modification is required by
  stations (bridge is "transparent")
• Both LANs can be transmitting simultaneously
  without collision
• Filtering is necessary if bridge repeats all
  traffic, then both LANs will see combined
  traffic, causing bad performance (like a large
  LAN)
• Performance depends on most traffic staying
  intra-LAN

9
Transparent Learning Bridges (cont)

• Bridge functions
• Stores frames from LAN and forwards as necessary
• Learns station addresses when stations transmit a
  frame
• Resolves possible loops in topology through a
  spanning tree algorithm
• Spanning tree is used to broadcast frames when
  destination station has unknown location

10
Transparent Learning Bridges (cont)

• Contains routing table (or forwarding database)

11
Transparent Learning Bridges (cont)

• Associates an outgoing port for each known
  destination station
• When frame is received, the dest. address is
  looked up in database
• If frame arrives on same port as in database,
  frame is discarded means already on correct LAN
• If frame arrives on different port, frame is
  forwarded to specified port
• If address is not found, frame is transmitted on
  ports determined by spanning tree algorithm
  except the port where it was received

12
Transparent Learning Bridges (cont)

• Learning algorithm to update routing directory
• For every received frame, bridge records incoming
  port and source address
• Finds source address in database
• If not there, it is added with incoming port
• If there with different port , then port is
  updated
• If there with same port , no change
• In all cases, start a timer
• When time-out, then entry is assumed out-of-date
  and removed

13
Transparent Learning Bridges (cont)

• Spanning tree algorithm to resolve loops (if
  loops are possible)
• Loops can cause problem in learning algorithm

14
Transparent Learning Bridges (cont)

• Station Y sends a frame to station X through two
  bridges at time T0
• Bridge A forwards frame to LAN 1, then bridge B
  forwards its frame later
• Station X gets two copies of frame
• When bridge A forwards frame to LAN 1, bridge B
  will think station Y is on LAN 1 (frame contains
  source and dest. addresses)
• Later, bridge A will think the same when bridge B
  forwards frame

15
Transparent Learning Bridges (cont)

• By spanning tree algorithm, bridges decide on a
  spanning tree to prevent loops

16
Wrap Up

• Bridges allow transparent interconnection of LAN
  segments
• Work at MAC layer
• Transparent learning bridges automatically build
  routing tables by learning locations of stations
  when they transmit
• Spanning tree is used to broadcast frames to
  stations with unknown locations

17
Spring 2006 EE 5304/EETS 7304 Internet Protocols
Lecture 4
Network layer, routing considerations
Tom Oh Dept of Electrical Engineering taehwan_at_engr
.smu.edu
18
Outline

• Network layer
• Routing considerations

19
Network Layer (OSI Layer 3)
- hides details of network from applications -
end to end reliability, flow control
Transport
Layer 4
- routing, congestion control, connection control
Network
Layer 3
- hides details of physical layer from network -
adds reliability, synchronization, flow control
Data link
Layer 2
- unguaranteed, unstructured bitstream
transmission
Physical
Layer 1
20
Network Layer (cont)

• Network layer uses services of data link layer
  (reliable point-to-point transmission between
  nodes) and adds packet switching for end-to-end
  connectivity
• Provides end-to-end delivery of packets as
  service to higher transport layer
• Can offer connectionless or connection-oriented
  service
• Establishes and manages connections through
  subnet (if connection-oriented)

21
Network Layer (cont)

• Determines routes for packets from sender to
  destination
• Exercises congestion control to maintain network
  efficiency during heavy traffic loads
• Allows transport layer to send data from host to
  host without need to know network details
• Unlike other layers, nodes need to share
  information among themselves to make decisions
• Routing protocol adds complexity

22
Network Layer - Issues

• Routing how to select the best route through the
  network
• Congestion control global problem of too much
  traffic for limited resources (vs flow control)
• Addressing how to identify hosts and nodes
• Internetworking how to deliver packets across
  network of different networks

23
Routing Considerations

• Routing determines a path through the network
  when more than one path is possible
• Sometimes no choice, e.g., no routing in
  hierarchical topologies (eg, trees)
• 1. Connection-oriented or connectionless?
• Connection-oriented
• Requires computation and set-up of path (virtual
  circuit) before any packets are sent
• Virtual" because circuit is not reserved
  packets from different virtual circuits can share
  same path and same physical links

24
Routing (cont)

• During set-up, hosts can specify requirements,
  eg, bandwidth, max. delay
• Network has chance to reject (block) connection
  or reserve resources for connection
• Each node along route knows next node to forward
  packets
• Packets of same virtual circuit maintain their
  sequential order
• Vulnerable to node/link failures

25
Routing (cont)

• Connectionless
• Path for each packet (datagram) is computed
  independently
• No set-up or disconnect phases
• No reservations, so usually "best-effort" type of
  network service
• More processing nodes make a routing decision
  for each datagram
• Datagrams may arrive at destination node out of
  original sequence
• More reliable and adaptive to failures

26
Routing (cont)

• 2. Routing algorithm to select best route
  according to some criteria
• Requires information from other nodes that's
  exchanged cooperatively
• Complicated by failures and congestion
• Examples of routing algorithms flooding,
  deflection routing, random routing, source
  routing, least cost (shortest distance) routing

27
Routing (cont)

• 3. Issues
• Where are decisions made? each node (hop-by-hop
  routing), central node (centralized routing), or
  source node (source routing)
• Is decision made per packet (datagram) or per
  session (virtual circuit)?
• Static (routing info. is changed only for
  topology changes) or adaptive (routing info. is
  changed continuously or periodically) or random?

28
Routing (cont)

• Issues (cont)
• How much information is available? none, local
  (each node uses own info.), adjacent nodes, or
  all nodes (centralized)
• What are performance criteria (routing metric)?
  number of hops, cost per link, delay,
  throughput,...

29
Routing (cont)

• 3. Routing tables
• Virtual circuit (connection-oriented)
• Routing table is updated during connection set-up
  phase
• All packets of same virtual circuit follow same
  route
• Routing table incoming link and virtual circuit
  ID, outgoing link and virtual circuit ID
• Why translate virtual circuit ID? global IDs
  will limit number of connections, and time
  consuming to find unused ID
• Advantages no routing decision per packet
  packets maintain sequential order
• Disadvantages link or node failure can bring
  down virtual circuit

30
Routing (cont)

• Example

IN
OUT
link
VC
link
VC
in-link1
X1 X2
out-link5 out-link2
Y1 Y2
in-link2
W1 W2
out-link4 out-link8
Z1 Z2
31
Routing (cont)

• Datagrams (connectionless)
• Route of each packet is decided independently
• Routing table destination, outgoing link or
  next node, cost
• For given destination, choose outgoing link based
  on some cost measure, eg, distance
• Advantages adaptability to changing network
  conditions (e.g., link or node failures,
  congestion)
• Disadvantages more processing per packet
  sequential order of packets is not maintained

32
Routing (cont)

• Example

Dest.
Outgoing link
Cost
A
outlink1 outlink2
X1 X2
B
outlink1 outlink2
Y1 Y2
33
Routing (cont)

• 4. Routing protocol (or update algorithm)
  protocol to share information to update tables
• Is routing static or dynamic?
• Update time more frequent is more accurate, but
  more overhead
• Consistency keep accurate and same information
  at all nodes

34
Routing (cont)

• 5. Levels of routing
• In large networks (internets), packets may travel
  within subnets and between multiple subnets
• Autonomous system (TCP/IP terms) or routing
  domains (OSI terms) subnetwork or group of
  subnetworks under single administration
• Nodes in same routing domain are interior routers
  or gateways (TCP/IP) or intradomain intermediate
  systems - ISs (OSI)
• Each has own interior gateway protocol (TCP/IP)
  or intradomain IS-IS protocol (OSI)

35
Routing (cont)

• Nodes exchange routing information and make
  routing decisions-- protocol is determined by
  administration
• RIP (routing information protocol) earlier and
  OSPF (open shortest path first) are widely used
• Autonomous systems are connected through exterior
  routers or gateways (TCP/IP terms) or interdomain
  ISs (OSI terms)
• Must conform to a standard protocol for
  interconnectivity, eg, BGP (border gateway
  protocol)

36
Layered Routing
Inter-AS routing
Autonomous systems
Intra-AS routing
37
Spring 2006 EE 5304/EETS 7304 Internet Protocols
Routing algorithms
Tom Oh Dept of Electrical Engineering taehwan_at_engr
.smu.edu
38
Outline

• Static routing (Comer Pg211, 399 and 400)
• Source routing
• Flooding (Comer Pg. 14)
• Dijkstras algorithm
• Bellman-Ford algorithm

39
Static Routing

• Routes are computed once and programmed into
  switches routing tables
• Routes do not change
• Suitable if network topology and traffic patterns
  do not change much

40
Source Routing

• Sender decides on route for each packet
• Route is specified within packet header
• Switches read packet header and forward packet
  along route
• Suitable when sender wants a certain route (e.g.,
  for testing) or avoid other routes (e.g.,
  competitors networks)
• Costs
• Extra fields in packet header
• Processing burden on sender to discover routes

41
Flooding

• Flooding is used to broadcast topology changes to
  all nodes or when exact location of destination
  is unknown
• Advantages
• Simple
• Needs no network information or routing tables
• Robust for failure-prone networks
• Shortest path is always found

42
Flooding (cont)

• Source node broadcasts packets to all neighbors,
  these broadcast to their neighbors, etc.
• Various methods to prevent infinite number of
  packets
• Node does not broadcast packet back on incoming
  link
• Packet has unique ID (eg, source node number,
  sequence number)
• If a node receives a duplicate copy, it discards
  the packet
• Packet keeps hop count
• Discarded when hop count reaches a limit

43
Flooding (cont)

• Alternatively, broadcast only along spanning tree
• Spanning tree is subset of connectivity graph
  that connects all nodes with no cycles
• Spanning tree must be maintained and updated
  somehow

44
Flooding (cont)
2
2
1
1
3
3
3
2
2
3
A
A
3
2
1
1
3
3
2
2
Flooding by broadcasting
Flooding along spanning tree
45
Random routing

• Outgoing link for a packet is chosen randomly
  according to set of probabilities
• Probabilities could be based on link rate or
  other fixed link info. to distribute traffic
  uniformly
• Simple and requires no network information, but
  routes are sub-optimal and looping must be
  prevented

46
Deflection (Hot Potato) Routing

• Objective is get rid of packets as quickly as
  possible, based on only local information
• Put outgoing packet on shortest output queue
• Minimizes chance of buffer overflow, but packet
  may not get closer to destination node
• Can be combined with static routing
• Each packet has assigned link (according to table
  lookup)
• If more than one packet are contending for same
  link, one succeeds, others are deflected

47
Least Cost (or Shortest Distance) Routing

• Assume knowledge of entire network
• Network is represented by graph of nodes and
  links
• Each link has assigned length or more generally
  "cost" (eg, function of distance, capacity, load,
  delays, etc.)
• Objective is find route between
  source-destination nodes with smallest total
  length or least cost
• 2 widely used algorithms
• Dijkstra (link-state routing protocols)
• Bellman-Ford (vector-distance routing protocols)

48
Least Cost Routing (cont)
2
2
4
4
3
6
3
A
A
2
2
4
1
1
1
1
3
3
Network graph
Spanning tree representing least-cost routes from
node A to all other nodes
49
Dijkstras Algorithm

• Each node has label (X,Y) where X is previous
  node and Y is distance from source node along
  best known path (up to that time)
• Label is either tentative or permanent
• Becomes permanent when it represents the shortest
  possible path from source node
• Algorithm proceeds to find shortest paths from
  source node to all other nodes, in increasing
  order of distance

50
Dijkstras Algorithm (cont)

• Proceeds in stages
• By stage k, will have found k shortest paths from
  source node
• These nodes are included in set M
• At next stage, shortest path is found to a node
  not in M, this node is added to M
• Ends when M includes all nodes
• Result is shortest path spanning tree with root
  at source node

51
Dijkstras Algorithm (cont)

• Notation
• 1 source node
• cij distance (cost) between nodes i and j
• M set of permanent nodes found so far
• (X,Y) node label
• X previous node
• Y distance from source node to this node

52
Initialize all node labels to (-,8). Add node 1
to set M M1
From last node added to M, update the labels at
neighboring nodes if shorter distance can be
found, change node label to shortest route.
Until M includes all nodes
Among the nodes not currently in M, choose node
with smallest distance in its label. Make
this node permanent and add to M.
53
Dijkstras Algorithm - Example Graph
5
2
3
3
5
2
3
1
6
1
2
2
1
1
4
5
54
Initialize node labels. Add node 1 to set M
M1.
(-,8)
(-,8)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(-,8)
(-,8)
54
55
Update labels at neighboring nodes from node 1.
(1,2)
(1,5)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(-,8)
56
Choose node 4 to add to M M1,4. Make 4
permanent.
(1,2)
(1,5)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(-,8)
57
Update labels at neighboring nodes from node 4.
(1,2)
(4,4)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
58
Choose 2 (or 4) to add to M M1,4,2. Make 2
permanent.
(1,2)
(4,4)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
59
From 2, update labels at neighboring nodes.
(1,2)
(4,4)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
59
60
Choose 5 to add to M M1,4,2,5. Make 5
permanent.
(1,2)
(4,4)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
61
From 5, update labels at neighboring nodes.
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
62
Choose 3 to add to M M1,4,2,5,3. Make 3
permanent.
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
63
From 3, update label at node 6.
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
64
Add 6 to M M1,4,2,5,3,6. Make 6 permanent.
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
65
Spanning tree representing least cost routes
(1,2)
(5,3)
2
3
(5,4)
(-,0)
1
6
4
5
(1,1)
(4,2)
66
Dijkstras Algorithm - Table Form
M
Node 2
Node 3
Node 4
Node 5
Node 6
1
(1,2)
(1,5)
(1,1)
(-,8)
(-,8)
1,4
(4,4)
(4,2)
(1,2)
(1,1)
(-,8)
1,4,2
(4,4)
(1,2)
(1,1)
(4,2)
(-,8)
1,4,2,5
(5,3)
(5,4)
(1,2)
(1,1)
(4,2)
1,4,2,5,3
(5,3)
(5,4)
(1,2)
(1,1)
(4,2)
1,4,2,5,3,6
(5,3)
(5,4)
(1,2)
(1,1)
(4,2)
67
Bellman-Ford Algorithm

• Bellman's optimality principle if path is
  optimal, any segment (between any two nodes)
  along this path must be optimal between those two
  nodes
• Algorithm proceeds to find shortest paths from
  source node to all nodes, in increasing order of
  number of hops
• Find shortest path of 1 hop (max.), then shortest
  path of 2 hops (max.), etc.
• Stop when next iteration does not change anything

68
Bellman-Ford Algorithm (cont)

• Proceeds in iterations
• Each node is responsible for updating its own
  label in each iteration
• Node looks at routes through all of its neighbors
• Chooses neighbor with shortest route
• Ends when no more changes
• Result is shortest path spanning tree with root
  at source node

69
Initialize all node labels to (-,8) except source
node label is (-,0)
Each node calculates routes through all
neighbors.
Iterate until no more changes
Choose shortest route and update label if
necessary.
70
Example Initialize node labels.
(-,8)
(-,8)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(-,8)
(-,8)
71
Iteration 1 each node calculates routes through
all neighbors, chooses shortest route, updates
labels if necessary.
(1,2)
(1,5)
5
2
3
3
5
2
(-,8)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(-,8)
72
Iteration 2
(1,2)
(4,4)
5
2
3
3
5
2
(3,10)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
73
Iteration 3
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
74
Iteration 4 no change ? stop
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
75
Node labels define shortest path spanning tree
(same as before)
(1,2)
(5,3)
5
2
3
3
5
2
(5,4)
3
(-,0)
1
6
1
2
2
1
1
4
5
(1,1)
(4,2)
76
Bellman-Ford Algorithm - Table Form
iteration
Node 2
Node 3
Node 4
Node 5
Node 6
0
(-,8)
(-,8)
(-,8)
(-,8)
(-,8)
1
(1,5)
(1,2)
(1,1)
(-,8)
(-,8)
2
(4,4)
(1,2)
(1,1)
(4,2)
(3,10)
3
(5,3)
(5,4)
(1,2)
(1,1)
(4,2)
4
(5,3)
(5,4)
(1,2)
(1,1)
(4,2)