On looking at this again, it appears that BGP Multipath only works when the eBGP sessions are terminated on the same box.
The scenario here is two eBGP session to the same ISP, but terminating on two different customer edge routers (with an iBGP session between them). In the lab tests I've done, I can see the two entries in the BGP table (one learned via the directly connected eBGP neighbour and one learned through iBGP (from the other eBGP session on the other router), but only the best path (via the eBGP link) gets entered into the RIB. There is a command "maximum-paths eibgp" to load-share across eBGP and iBGP paths, but this is limited to connections within an MPLS VPN, and not the global routing table.. Maybe the outbound load-sharing will occur naturally through the IGP, and dynamic default routing (i.e. the closest exit point)? Alternatively, another possibility may be to arbitrarily prefer one link for some traffic, using AS paths or prefix lists together with NetFlow data, as a 'first cut' and then optimise over time. Cheers, Matt On 9 February 2010 05:52, Aftab Siddiqui <[email protected]> wrote: > hi Matthew, > > Keeping the current internet full feed in view its around 300k routes and > sup720-3BXL should support 1million routes (its cisco though :p). So even if > you terminate the links on 2 different edges coming from the same AS it > should work fine. > > If you are trying "bgp bestpath as-path multipath-relax" kindly share the > outcomes because in my opinion it is used to load share between different > as-path. I have never tried it before. > Regards, > > Aftab A. Siddiqui > > > On Tue, Feb 9, 2010 at 12:59 AM, Matthew Melbourne <[email protected]> > wrote: >> >> Thanks for the pointers towards eBGP Multipath. Can I check that this >> still >> works if two links are terminated on different edge routers (though with >> iBGP between the edge routers). I assume this will use additional TCAM >> resources (Sup720-3BXL) in maintaining two routes per prefix, which could >> be >> significant for a full BGP feed? >> >> Cheers, >> >> Matt >> >> -----Original Message----- >> From: Erik Cuevas [mailto:[email protected]] >> Sent: 05 February 2010 12:33 >> To: Matthew Melbourne >> Subject: RE: [c-nsp] Load-sharing with two links to the same ISP >> >> Did you check out BGP multipath? >> >> >> http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080094431 >> .shtml >> >> >> or is the AS Path is different try... >> >> bgp bestpath as-path multipath-relax(its hidden) >> >> -----Original Message----- >> From: [email protected] >> [mailto:[email protected]] On Behalf Of Matthew Melbourne >> Sent: Friday, February 05, 2010 6:33 AM >> To: [email protected] >> Subject: [c-nsp] Load-sharing with two links to the same ISP >> >> Hi, >> >> What techniques are available to load-share traffic on two links (of >> equal bandwidth) to the same ISP (same AS) given that BGP only enters >> the best path into the RIB? We could announce our prefixes over both >> links, but splitting the preferred path announcements over the two >> links, either using MED or ISP communities, but this only really >> addresses inbound traffic. More of an issue is trying to load-share >> outbound traffic; we assume we'll learn the same set of prefixes over >> both links from the same ISP - one technique may be to simple split >> the IPv4 address space in half and local-pref accordingly to prefer >> one link or the other depending on the destination IP prefix? >> >> Cheers, >> >> Matt >> >> -- >> Matthew Melbourne >> _______________________________________________ >> cisco-nsp mailing list [email protected] >> https://puck.nether.net/mailman/listinfo/cisco-nsp >> archive at http://puck.nether.net/pipermail/cisco-nsp/ >> >> No virus found in this incoming message. >> Checked by AVG - www.avg.com >> Version: 9.0.733 / Virus Database: 271.1.1/2669 - Release Date: 02/05/10 >> 07:35:00 >> >> _______________________________________________ >> cisco-nsp mailing list [email protected] >> https://puck.nether.net/mailman/listinfo/cisco-nsp >> archive at http://puck.nether.net/pipermail/cisco-nsp/ > > -- Matthew Melbourne _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
