Internetwork Routing

1
Internetwork Routing Consider 5 networks
connected by 6 multiprotocol routers.
Once a graph of the internetwork is created, the
distance vectors can be determined.
2
Internetwork Routing

• Within each network an interior gateway protocol
  is used. Between these networks, an exterior
  gateway protocol is used.
• Gateway is an older term for router.

3
Internetwork Routing

• A typical internet packet starts on a LAN
  addressed to the local multiprotocol router (in
  the MAC layer header). Then the network layer
  decides which multiprotocol router to forward the
  packet, using its own routing tables.
• If that router can be reached using the packets
  original network protocol, it is immediately
  forwarded there.
• Otherwise, the packet is tunneled to its
  destination.
• If a packet is too large for a particular
  gateway, it is chopped into smaller fragments. It
  is reattached after it travels through that
  gateway.

4
Tunneling Although many different types of
networks and protocols can be found on an
internetwork, there is a common special case
where the source and destination are on similar
networks, but there is a different network in
between.
5
Tunneling

• The sender constructs an IP packet to the
  destination, containing the destinations
  address.
• This packet is inserted into an Ethernet frame
  addressed to a multiprotocol router.
• The router removes the IP packet and inserts it
  into the payload field for a WAN network layer
  packet.
• The payload address field is the multiprotocol
  router closest to the destination. This router
  removed the IP packet from the payload, and then
  sends it to the destination.
• The WAN acts like a tunnel between points.

6

• Internet Control Protocols
• The operation of the internet is mostly
  controlled by routers (no surprise). There are 5
  major protocols
• ICMP
• ARP
• RARP
• OSPF
• BGP

7
Internet Control Message Protocol (ICMP) When an
unexpected event occurs at a router, the event is
reported by the ICMP. ICMP is also used to test
the internet. Some of these messages are shown
below.
8
Internet Control Message Protocol (ICMP)

• Destination Unreachable - message occurs when the
  subnet or router cannot locate a destination, or
  when a packet set for DF cannot be delivered (a
  small packet network blocks the path).
• Time Exceeded message - means that the Time To
  Live field has reached zero.
• Parameter Problem - indicates an illegal value in
  the header field.
• Redirect - tells a sending host that a packet is
  being incorrectly routed.

9

• Address Resolution Protocol (ARP)
• Maintaining router tables is a considerable
  chore. When an address is encountered for a
  destination that is unknown to a router, the
  router can broadcast a query to all routers about
  that address. This is performed using ARP.
• Reverse Address Resolution Protocol (RARP)
• This allows a newly-booted machine to determine
  its corresponding IP address. Sometimes an
  associated protocol, BOOTP is also used.

10

• Interior Gateway Routing Protocol (OSPF)
• This protocol handles routing inside (interior) a
  local network (gateway). The most current
  protocol for this is Open Shortest Path First
  (OSPF). OSPF supports 3 kinds of networks
  connections
• Point-to-point lines between exactly 2 routers
• Multi-access networks with broadcasting (most
  LANs)
• Multi-access networks without broadcasting (most
  packet-switched WANs)

11
Interior Gateway Routing Protocol (OSPF)
Within each local area, every router has the same
link state database and also uses the same
shortest path algorithm.
12
Interior Gateway Routing Protocol (OSPF)

• OSPF categorizes routers into 4 classes
• Internal routers wholly inside one area
• Area border routers that connect 2 or more areas.
• Backbone routers
• AS boundary routers talk to routers in other
  ASes.

13
Interior Gateway Routing Protocol (OSPF)

• OSPF exchanges information with adjacent routers.
  One router is chosen to be the designated
  router. It is then considered to be adjacent to
  all other routers on the LAN, and exchanges
  information with them.
• During normal operation, each router periodically
  floods adjacent routers with LINK STATE UPDATE
  messages. These flooded messages are acknowledged
  by all adjacent routers.

• By analyzing the responses, a router can select
  the best adjacent router to send data to.

14

• Exterior Gateway Routing Protocol (BGP)
• Between autonomous systems (other networks), a
  different protocol, BGP, is used. To a BGP
  router, the world consists entirely of other BGP
  routers. They share a common network. There are 3
  BGP subnetworks
• Stub networks
• Multi-connected networks
• Transit networks

15
Exterior Gateway Routing Protocol (BGP)

• Stub networks - are dead ends and have only one
  connection to a BGP network.
• Multiconnected networks - could be used for
  transit traffic, but they REFUSE transit traffic.
• Transit networks - such as backbones, are able
  and willing to handle 3rd party packets.
• BGP is primarily a distance vector protocol. It
  records the line cost (time hops) to each
  destination. It also keeps track of the exact
  path used. Instead of telling each neighbor its
  cost data, it tells each neighbor the path it
  uses to each neighbor.

16
Example of BGP Router-to-Router Messages
17
Routing Principles Most packet transmissions
will require multiple hops across the subnet.
Routing algorithms fall into one of 2 categories
1) Nonadaptive algorithms 2) Adaptive
algorithms
18
Routing Principles
Nonadaptive algorithms compute the path between 2
points in advance. All routers located within a
network are provided with this information when
the system is booted up. Adaptive algorithms
change their routing paths due to constant
testing of current topology and traffic
congestion. They obtain their information by
monitoring other nearby routers.
19
Routing Principles
Distance Vector Routing Most networks use
adaptive (dynamic) routing methods. One of the
two most popular methods is distance vector
routing. In this method, each router maintains a
table (vector) giving the best-known distance to
each destination and which line to use to get
there. These tables are updated by exchanging
information with its neighbors. Each router
contains a routing table containing an entry for
every other router in the subnet.
20
Routing Principles

• Each entry contains
• the preferred outgoing line to use for that
  destination
• an estimate of the time or distance to that
  destination
• The distance is not really distance. In
  this case, distance means the number of hops, or
  the queue length, or the delay time.

21
Routing Principles

• Once every so often ( milliseconds ), each router
  sends to each neighbor a list of estimated delays
  to each destination. It also receives similar
  lists from its neighbors.
• The router then calculates the new distance to
  each neighbor and update its routing table
  accordingly.

22
Routing Principles
In the figure shown above, a router would
calculate the distance between J to G along
several paths. It will quickly determine that the
fastest path is from J to H to G.
23
Routing Principles

• Link State Routing
• The other type of dynamic routing is link state
  routing. Using this method, each router will
• discover its neighbors network addresses
• measure the delay to each of its neighbors
• construct a packet containing this information
• send this packet to other routers
• compute the shortest path to each router

24
Routing Principles
When a router is booted, it first learns about
its neighbors by sending a special HELLO packet
on each point-to-point line. Other routers on the
line will respond by identifying themselves. The
identifiers must be globally unique.
25
Routing Principles

• Measuring the line cost (delay) requires a
  choice. Do you measure the delay when the line is
  not busy ? Or do you measure the line when it is
  heavily congested ?
• The first inclination is to measure during
  congestion. After all, congestion is a legitimate
  factor.
• Remember that the line cost information will
  eventually end up in the router tables. Consider
  the following subnet.

26
Routing Principles
Suppose the line from C to F is heavily
congested, but the line from E to I is lightly
trafficked. If the line cost includes congestion,
then all of the routers will update their tables
to favor path E to I. Suddenly, E to I will
become congested, while C to F will become
uncongested. This problem will repeat itself over
and over again.
27
Routing Principles
Once the neighbors are located and the line costs
have been calculated, the next step is for a
router to create a packet containing this
information. The packet begins with the identity
of the sender. It then adds a sequence number,
its age, a list of neighbors, and the delay to
each neighbor.
28
Routing Principles

• One unanswered question is when to build these
  link state packets. They could be built
  periodically, or built after a major event has
  occurred.
• The link packets are then distributed by
  flooding. If a router receives a link state
  packet, it checks the sequence number. If the
  sequence number is higher than the last sequence
  number from the sender, the router keeps the
  packet and updates its tables accordingly.

29
Routing Principles

• If a router goes down or fails, when it is booted
  up again its sequence number will start at zero.
  When it floods its neighbors, the neighbors will
  ignore it since the sequence number is lower than
  the last one.
• To prevent this event from happening, we must
  return to the age part of a link state packet.
  When a packet is accepted by a router, it is
  given an age (say 10 minutes). After each minute,
  the age is decremented. When the age reaches
  zero, the data in the table is updated by the
  next packet, regardless of the sequence number.

30
Routing Principles
Hierarchical Routing As networks grow, so does
the number of routers. And so also does the size
and complexity of routing tables. Eventually, it
is no longer practical for every router to have
an entry for every other router. Instead, a
router hierarchy can be developed. With
hierarchical routing, the routers are divided
into regions. Each router knows how to route
packets to every other router in its region. But
it knows nothing outside of its region.
31
Routing Principles
Two-Level Hierarchy with Five Regions
32
CONGESTION CONTROL
When too many packets are present on a part of
the subnet, congestion can occur. This severely
limits the amount of packets delivered.
33
CONGESTION CONTROL

• If a stream of packets suddenly arrives on
  several input lines, and all need the same output
  line, a queue will develop.
• If there is insufficient memory, packets will be
  lost.
• If more memory is added, the problem becomes
  worse. The size of the queue increases, but
  eventually transmitter machines time out without
  receiving an acknowledgement.
• The transmitters response to time outs is to
  send the packet again. The router memory will
  then become clogged with multiple copies of
  packets.

34
CONGESTION CONTROL

• Slow routers can also cause congestion. Queues
  continue to get larger. Again we have time out
  problems.
• Congestion control deals with ensuring that the
  subnet is able to carry the desired traffic. It
  is a global issue and involves ALL hosts,
  routers, and store-and-forward processes within
  routers.
• Flow control, however, is a data link issue. It
  is concerned with point-to-point traffic between
  a given sender and receiver.
• Flow control requires direct feedback from the
  receiver. Congestion control does not.

35
CONGESTION CONTROL

• General Principles of Congestion Control
• There are 2 approaches to congestion control
• Open loop control (no feedback)
• Closed loop control (feedback)
• Open loop control consists of methods such as
  deciding which packets a router accepts and which
  packets a router discards. The current state of
  the network is not an issue for open loop control.

36
CONGESTION CONTROL

• Closed loop control uses feedback. It is
  therefore necessary to
• Monitor the subnet and detect when congestion
  occurs.
• Pass this information to places where action can
  be taken.
• Adjust system operation to correct the problem.

37
CONGESTION CONTROL

• One method for monitoring congestion is to
  calculate the percentage of all packets that are
  discarded due to
• lack of buffer memory space
• the average queue length
• the number of packets that timeout and are
  retransmitted,
• the average packet delay.
• If the capacity cannot be increased by adding new
  lines, then the only alternative to correct
  congestion is to decrease the load by either
  denying access to some users or by degrading
  service to some users.

38
Policies that Affect Congestion
39
CONGESTION CONTROL

• Traffic Shaping
• Congestion often occurs in large bursts of
  traffic. If hosts transmitted at a uniform rate,
  much congestion could be prevented. An open loop
  method for congestion management is called
  traffic shaping.
• Traffic shaping regulates the average rate of
  data transmission.
• It is an agreement made when a virtual circuit is
  set up.