nanog mailing list archives

Re: MD5 BGP performance on a VXR?


From: Henning Brauer <hb-nanog () bsws de>
Date: Fri, 11 Jun 2004 21:13:16 +0200


* Patrick W.Gilmore <patrick () ianai net> [2004-06-11 20:54]:
On Jun 11, 2004, at 8:21 AM, Newell, Tony wrote:
My first question would be how big is your prefix list per BGP session?
What is really going to task this router with 25 sessions is the BGP
Scanner and BGP Router processes.  To my knowledge MD5 is just for
authenticating the session.  I could be wrong.
Every TCP packet in the BGP session (including HELLOs) will have to go 
through the MD5 process.

there is no HELLO in bgp. and it is not really related to bgp either, 
it is just the common case that they're used together. with tcp md5sig, 
each and every packet gets a md5 signature - build from the packet header 
and a shared secret - added, and the receiving side - which, of course, 
has to have the secret for that - does the same again. if the signature 
in the packet and the signature the receiver calculated don't match, 
the packet is discarded (well, should. FreeBSD's implementation does 
sign outgoing packets and simply ignores signatures on incoming 
packets, very useful. ok, I don't know wether this has been fixed, but 
thanks for the laugh).

This happens even if things like the sequence number is wrong (at least 
on some versions of IOS).

I consider this Yet Another IOS Bug.

-- 
Henning Brauer, BS Web Services, http://bsws.de
hb () bsws de - henning () openbsd org
Unix is very simple, but it takes a genius to understand the simplicity.
(Dennis Ritchie)


Current thread: