Date -  08/08/'03

Scribe for Lecture - 6

Sudheer V.V.N.


Topics :

1) Clarification about BGP

2) Routing Instability

 

 

How BGP is implemented over TCP?

- Actually   TCP   run   over   IP   where   the   actual   routing   for   data   takes   place   but   if   we   come  to   BGP   it   routes   the   control

 flow, the control flow   is   implemented   on   TCP


BGP ;

 BGP is used for routing in between ASes (Externel Gateway Protocol). It is a path vector routing protocol

It will do its routing using messages

- a Open message is for intialization during which it exchanges its AS numbers

- a Update message is that which is actually used for routing there are 2 in it

  -- a Announcement is raised if there is a new link or or a new node

  -- a Withdrawl when a link or a node goes down

    when a router receives this messages then it will update its routing table and forward that

- a Notification  message to close a connection

- a Keep-Alive to tell the peer router that the router is alive

----------------------------------------------------------------------------------------------------------------------------

BGP uses Incremental approach to update state changes

- This   will   respond   only   when   there   is   a   change  of   state  but   for  this   we  need  to  maintain  per  client state  which is  a  overhead. In

 this    some    errors   may   arise

----------------------------------------------------------------------------------------------------------------------------

Some of the Path Attributes are

Multi Exit Descriminators (MED) : These   are   used   when   there   are   more   than   one route to a particular destination AS having different links

 in between then the link which is best is set in this attribute.

LOCAL PREF : If we can go to to a an AS via 2 ASes then this AS preference is set in this Attribute

----------------------------------------------------------------------------------------------------------------------------

For BGP the rules for the order of selection of routes are

- It will check the LOCAL- PREF attribute if it is not set then it will go for AS-PATH followed by MED

     Then it will see wheter it is E-BGP or I-BGP The preference of E-BGP > I-BGP

----------------------------------------------------------------------------------------------------------------------------

BGP takes much to converge i.e. if there is a change in topology then it takes time to all router come to know that and change their routing tables this

 may take even few minutes to converge


Instability in routing :

 What is instability ?

 - Instability is change in routes/topology , due to links/routers going down or new one comes 

Effects of Instability

- Lost packets - packets may lose due to change in routes

- Load on Routers - this is to update the routing tables continuously which causes cpu wastage

----------------------------------------------------------------------------------------------------------------------------

In IGP Instability is not of interest because

- smaller in size : as the size is small there is less updates

- single protocol is followed by all routers

----------------------------------------------------------------------------------------------------------------------------

Instability in BGP :

Instability is actually due to updates

                         UPDATES  can be divided in to

              1)Announcements     and         2) Withdrawls

                                     Withdrawls   are divided into 

                Implicit(By changing the value)      Explicit(by discarding the value)

 

Update can cause instability or some pathological (Redundant or which are not required ) update may come

This instability can be Forwarding Instability(topology changes) or Policy fluctuation(change of policies regularly)

----------------------------------------------------------------------------------------------------------------------------

Goals and Methodology

- Goal:  How many updates are comming  ?  among which How many are valid ?

- Method : 

 --- Collect BGP update traces at large exchange points

         Exchange point :: An exchange point is that in a room there will be a router of each AS which exchanges data through that exchange point the data is exchanged between them.

    This is public when there are more  than two routers of different AS                                         

    This is public when there  only two routers that exchange data                 

                                             

----------------------------------------------------------------------------------------------------------------------------

Default-free route is that when there is no route specified with the address present in that packet the it is sent to a route which is default for that router

----------------------------------------------------------------------------------------------------------------------------

What are the effects of instability on BGP

Route caching will not be effective:: i.e. if there is continuous change of route then caching that route for that destination may not be effective it has to be changed frequently

"Route flap storm" ::  when a router cannot even process Keep alive messages then it will think that router is failed and deletes from its routing table this continues

- Large state -dump :: when a line goes down and come up again it has to send all its routing table to the peer router this will dump large data on to the network.

----------------------------------------------------------------------------------------------------------------------------

Possible ways to reduce Instability :

1. Address aggregation

2. Route servers

3. Route Dampening

----------------------------------------------------------------------------------------------------------------------------

Address Aggregation ::

- This is combining the prefixes of two networks and send it to whole network(core of internet)

for example :

If there are networks with addresses 1.2.3.0/19 and 1.2.19.0/19 then these two addresses can be aggregated and can send it to the Core of internet the aggregated address is 1.2.3.0/18 in which both if them comes

But Aggregation is not possible in BGP

-- Multi homing- when a network is connected to more than one network the it is not possible to aggreagate with one

-- While changing it's connection from one network to another it may not change its address then also it is not possible to aggregate