Interface Output buffer and BGP

Unanswered Question
Apr 14th, 2010
User Badges:

Hi


Im looking at setting up a sites resilience using dual connections to two routers running BGP. If the two links are not equal i.e. the primary link that all traffic routes over in a normal situation is 100Mbps but the back link that will be used should the primary go down is only 10Mbps im trying to test how stable BGP will be in this situation should the 10Mbps link be "maxed out for long durations". trying to ram 30Mbps down the 10Mbps link I could get packet loss over the link but BGP remained stable "this is good" but was wondering if this is just luck. Does by default Cisco routers give higher priority in the output buffer to protocols such as BGP. Will other traffic always be dropped before BGP.


Any advice appreciated.


Kind Regards


Kev

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (1 ratings)
Loading.
kjbarrass Wed, 04/14/2010 - 02:36
User Badges:

Sorry I should have said output Queue not buffer. Been up way too early today



Cheers

Kev

Giuseppe Larosa Wed, 04/14/2010 - 03:55
User Badges:
  • Super Silver, 17500 points or more
  • Hall of Fame,

    Founding Member

Hello Kev,


a system hidden queue for routing protocol messages exist on several Cisco router platforms including ISR routers.


This is the reason behind the max-bandwidth command that decides how much of stated bandwidth can be assigned to user defined traffic classes and that defaults to 75%.


To be noted some of high end Cisco routers like C7500, GSR, and likely C7200 haven't this hidden system queue and it is left to the network engineer to configure a queue to protect routing protocols messages from being dropped outbound.


On BGP route reflector servers it makes sense to tune interface buffers and input queue in order to support many BGP sessions but this is probably not your case.


Hope to help

Giuseppe

kjbarrass Wed, 04/14/2010 - 04:02
User Badges:

Hi


This makes sense the core device is a 4507 and the remote router is a 2851 all running FIFO. I guess I was lucky in the lab that BGP didnt fail. Will do some further lab testing and look into other queueing methods if required.


Regards


Kev

Actions

This Discussion