[c-nsp] BGP soft-reconfiguration inbound impact

[Phil Mayers]

Frederic LOUI wrote:
> Hi everyone,
> 
> I spent some times googling/searching the mailing list but I could not 
> find any clear answer regarding
> memory impact related to "soft-reconfiguration inbound" statement. (If 
> you have any link/pointer, I'm interested !)
> 
> We're running a bunch of 760X (RSP7203CXL + 8x10G withc DFC3CXL) having 
> full BGP feed and some of them start to be really overloaded.
> 
> I've read now and then that it is not recommended to use 
> "soft-reconfiguration inbound" due to the extra memory used, but :
> 
> 1) What is the clear impact of this command ? (Is there an algorithm  
> formula O(n) that would help us to quantity the memory used ?)
> 2) Is this extra-memory still an issue with modern hardware ?

It depends on how many routes you have I think. If you've got the full 
feed, then I'd say you're going to pay a heavy price for soft-reconfig.

What does "sh ip bgp summ" say?

> 3) What is the common best practice ?

Modern BGP implementations tend to support route refresh, where you 
request the peer re-send it's RIB. There's no config needed for this - 
just to a "sh ip bgp nei" and see if refresh is available:

BGP neighbor is X  remote AS 64580, internal link
  Inherits from template iBGP-world for session parameters
   BGP version 4, remote router ID X
   BGP state = Established, up for 4w4d
   Last read 00:00:43, last write 00:00:31, hold time is 180, keepalive 
interval is 60 seconds
   Neighbor capabilities:
     Route refresh: advertised and received(new)

Of course, refreshing the full feed will take a while! There are I 
believe optimisations for this - ORF rings a bell - but since I don't 
deal in full-feeds, it's not something I'm up to speed with.

> 4) Are you using "soft-reconfiguration inbound", if yes how ? (i.e: Only 
> for troubleshooting purpose, "always on" as part of your configuration 
> template etc.)

We are because it's convenient to be able to do "sh ip bgp nei X 
received-routes" but we've got a very small routing table:

ac-core#sh ip bgp summary
BGP router identifier X local AS number 64580
BGP table version is 8375, main routing table version 8375
903 network entries using 105651 bytes of memory
1763 path entries using 91676 bytes of memory
153/13 BGP path/bestpath attribute entries using 21420 bytes of memory
20 BGP rrinfo entries using 480 bytes of memory
8 BGP AS-PATH entries using 192 bytes of memory
5 BGP community entries using 120 bytes of memory
11 BGP extended community entries using 264 bytes of memory
0 BGP route-map cache entries using 0 bytes of memory
0 BGP filter-list cache entries using 0 bytes of memory
BGP using 219803 total bytes of memory

Obviously that's not the case for you.