High router CPU utilisation on slow script about exabgp HOT 4 CLOSED

From thomas.mangin on January 22, 2013 10:57:51

Hello James,

It could be that as the script is very slow, the buffer on the TCP connection then fill up causing the kernel to put back pressure on the TCP connection (some BGP implementation even set the TCP window to zero to stop message being sent).

I never saw this issue with other BGP implementation but it could be an issue specific to the IOS train. I would advise not processing the routes on your script immediately but just buffer them and process them once things are quieter. Having a non-blocking read on the STDIN and processing only a few routes each time the loop is running may be a way.

I could buffer ExaBGP side but it would need to be an option, and would require a new command from the script to say "yep, send me more". If this solution does not work, I will consider this solution but I would rather avoid having to do it.

Regards,

Thomas

from exabgp.

From thomas.mangin on January 28, 2013 02:02:59

If you defined a new peer-group, the router has to go through the entire RIB to recalculate the routes to announce. If you are using a existing peer group, then it may be easier on the CPU as the table should already have been calculated for your other peers.

The CPU usage will only be for the initial export, after than, the updates should creates marginal CPU work load.

Status: Started
Labels: -Type-Defect Type-Other

from exabgp.

From thomas.mangin on February 17, 2013 08:46:53

Following our mail exchange, could you confirm if this CPU usage is normal or not please ?

from exabgp.

From thomas.mangin on February 21, 2013 10:20:27

closing as the issue is most likely a slow CPU on the router (nothing to do with ExaBGP).

Status: WontFix

from exabgp.