[c-nsp] bgp maximum-paths

[Chris Gotstein]

Already running "bgp bestpath as-path multipath-relax"

---- ---- ---- ----
Chris Gotstein, Sr Network Engineer, UP Logon/Computer Connection UP
http://uplogon.com | +1 906 774 4847 | chris at uplogon.com

On 4/15/2010 12:22 PM, Anton Kapela wrote:
> 
> looking at:
> 
>> It's inbound i'm mostly concerned with.  I'm taking full routes from
>> both providers.  Paths to both a relatively equal, so the path with the
>> lowest ID is winning, causing it to be overloaded.
> 
> and:
> 
>> I've looked into the communities option, but only Qwest supports them, my other provider doesn't support communities at this time.  Though the Qwest link is the one i want to push more traffic to, so maybe i can just use their communities and see what happens.
> 
> give me the impression you're concerned with 'attracting' even loads of bits/sec inbound, and outbound on your links. Depending on the 'qualitative difference' between the two networks (i.e. is your 2nd upstream simply between you and $transit_free_networks? or are they a reseller of a reseller of a tier1?), you may be able to get away with 'more' ECMP behavior for a larger number of prefixes: "bgp bestpath as-path multipath-relax" 
> 
> this means that even with different as's in the path, your router will generate ECMP routes for any BGP route with the same number of AS's in the path. So, if qwest shows you an average as path length of 3.x to 90% of the table, and $other isp shows you an average of 4.x, prepend inbound from quest (make their paths look longer), and then gain the ECMP-ness that comes from mostly-same AS-paths. 
> 
> as others noted, l3/l4 hashing will make some aspects of troubleshooting "slightly annoying" -- and this command will also cause your router to chew a small bit more dram for the CEF table due to new NH's for many thousands of routes that previously had none. 
> 
> probably best to toss this on and clear ip bgp * soft sometime after most of your users are fast asleep.
> 
> -Tk