[c-nsp] BGP load balancing (outbound)

Andrew Tolstykh andrew at atfam.com
Tue May 18 12:06:40 EDT 2010


> Is this per-packet, or per-flow load balancing?

The command allows FIB to receive and install multiple ECMP prefixes from BGP RIB.
The actual load-balancing algorithm is independent from this process, so standard per interface "ip load-sharing" config should apply.

> How much memory usage would it increase assuming 2-3 full tables in memory
> already (around 400Mb usage, 1Gb RAM)?  Looks like you've got 2 defaults only.


I need to lab this up to be absolutely certain, but the BGP RIB memory consumption shouldn't change at all.
FIB/CEF memory consumption is likely to double (in my case I would expect it to go from 49MB to something close to 100MB).

HTH,
Andrew

On 18/05/2010, at 11:58 PM, Randy McAnally wrote:

> I can find very little info on 'bgp bestpath as-path multipath-relax'
> 
> Is this per-packet, or per-flow load balancing?
> 
> How much memory usage would it increase assuming 2-3 full tables in memory
> already (around 400Mb usage, 1Gb RAM)?  Looks like you've got 2 defaults only.
> 
> --
> Randy
> 
> ---------- Original Message -----------
> From: Andrew Tolstykh <andrew at atfam.com>
> To: amps <djamps at gmail.com>
> Cc: cisco-nsp at puck.nether.net
> Sent: Tue, 18 May 2010 16:29:52 +1000
> Subject: Re: [c-nsp] BGP load balancing (outbound)
> 
>> ECMP load balancing works with the hidden "bgp bestpath as-path 
>> multipath-relax" command applied (thanks Anton!)
>> 
>> router bgp XXX
>> maximum-paths 2
>> bgp bestpath as-path multipath-relax
>> 
>> sh ip route bgp
>> B*    0.0.0.0/0 [20/0] via 13.13.13.3, 00:05:59
>>                [20/0] via 12.12.12.2, 00:05:59
>> 
>> sho ip cef 0.0.0.0 0.0.0.0
>> 0.0.0.0/0
>>  nexthop 12.12.12.2 FastEthernet1/0
>>  nexthop 13.13.13.3 FastEthernet1/1
>> 
>> If you are after the per-AS load-sharing configuration, then you 
>> should probably flip route-map TRANSIT-1-INT permit 10 and 20 statements.
>> Also don't forget to add blank "route-map TRANSIT-2-IN permit 20" statement.
>> 
>> route-map TRANSIT-1-IN permit 10
>> match as-path 105
>> set local-preference 210
>> route-map TRANSIT-1-IN permit 20
>> 
>> route-map TRANSIT-2-IN permit 10
>> match as-path 106
>> set local-preference 210
>> route-map TRANSIT-2-IN permit 20
>> 
>> HTH,
>> Andrew
>> 
>> On 18/05/2010, at 2:42 AM, amps wrote:
>> 
>>> Let's say I wanted to send traffic to even AS's out peer 1 and odd AS's
> out peer 2.
>>> 
>>> Am I crazy?  (Seems better than using metrics based solely on static IP
> ranges...)
>>> 
>>> Would this work?
>>> 
>>> route-map TRANSIT-1-IN permit 10
>>> set as-path prepend last-as 1
>>> !
>>> route-map TRANSIT-1-IN permit 20
>>> match as-path 105
>>> set local-preference 210
>>> !
>>> 
>>> route-map TRANSIT-2-OUT permit 10
>>> set as-path prepend xxxxx
>>> !
>>> 
>>> route-map TRANSIT-2-IN permit 10
>>> match as-path 106
>>> set local-preference 210
>>> !
>>> 
>>> route-map TRANSIT-2-OUT permit 10
>>> !
>>> 
>>> ip as-path access-list 105 permit [02468]$
>>> 
>>> ip as-path access-list 106 permit [13579]$
>>> 
>>> 
>>> On Mon, May 17, 2010 at 10:59 AM, Anton Kapela <tkapela at gmail.com> wrote:
>>> On Mon, May 17, 2010 at 10:42 AM, Andrew Tolstykh <andrew at atfam.com> wrote:
>>>> Just a shot in the dark, but I don't see load balancing enabled in your
> configuration snippet:
>>>> 
>>>> router bgp xxxxx
>>> 
>>> This question comes up semi-often, perhaps there ought to be a FAQ? (I
>>> may have missed it).
>>> 
>>> For different-upstream AS's, ecmp will not normally occur. You'll need
>>> to adjust the bgp bestpath determination functions, else the different
>>> next-AS's will prevent ECMP forwarding from being pushed to the FIB.
>>> 
>>> http://puck.nether.net/pipermail/cisco-nsp/2005-November/025993.html
>>> 
>>> -Tk
>>> 
>> 
>> _______________________________________________
>> cisco-nsp mailing list  cisco-nsp at puck.nether.net
>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>> archive at http://puck.nether.net/pipermail/cisco-nsp/
> ------- End of Original Message -------
> 
> 




More information about the cisco-nsp mailing list