BGP and convergence time

Joel Jaeggli joelja at bogus.com
Thu May 20 04:34:05 UTC 2010



On 05/12/2010 02:41 PM, Scott Weeks wrote:
> 
> 
> --- danny at tcb.net wrote: From: Danny McPherson <danny at tcb.net> On May
> 12, 2010, at 9:40 AM, Jay Nakamura wrote:
> 
>> I just tested this and, yes, with Cisco to Cisco, changing the
>> setting won't reset the connection but you have to reset the
>> connection to have the value take effect.  I need to look up what
>> happens when two sides are set to different values and which one
>> takes precedent.
> 
> : The holdtime isn't technically negotiated, both sides convey their
>  : value in the open message and the lower of the two is used by both
>  : BGP speakers.
> 
> This isn't a negotiation?
> 
> 
> : IIRC, neither J or C reset the session with the timer change, but
> the : new holdtimer expiry value doesn't take effect until then.
> 
> We use Alcatel 7750s.  Damn thing just resets the session; no
> warning, no nothing. :-(
> 
> 
> 
> : One other thing to note is that by default, keepalive intervals in
>  : those implementations are {holdtime/3}.  Normally, if you're
> setting : holdtime to something really lower (e.g., 10 seconds) you
> might want : to increase the frequency of keepalives such that the
> probability of : getting one through in times of instability rise.
> In particular, : congestion incurred outside of BGP, as update
> messages themselves : will serve as implicit keepalives, and with the
> amount of churn in BGP, : empty updates (keepalives) are rare for
> most speakers with a global BGP : view.
> 
> I have been looking for info on the negative impact on a router by
> increasing the keepalive frequency to a high rate.  I'm sure it's
> minimal for a few BGP peers, but I could imagine with a lot of peers
> it's a non-zero impact.

with a keep alive interval of 10 seconds you can expect to get 10pps
from a 100 peers. the keepalive message is 19bytes

That doesn't seem particularly hurtful even by the standards of 5 year
old control plane processors.

> scott
> 
> 
> 
> 
> 
> 




More information about the NANOG mailing list