6500 BGP Load-Sharing Questions

Unanswered Question

I have "maximum-paths" configured and seeing two entries in the RIB for remote prefixes at the hub core 6500 (it has two EBGP connections to two CE from 2 carriers each).


The remote site has 2 CE routers connected to two carriers, one carrier each, and seeing the same as-path length for the hub routes. The remote LAN switch has two equal EIGRP routes for the hub prefixes.


However the remote is only using one carrier for inbound traffic, doesn't doing the load-sharing.


I think the problem could be due to hub 6500 CEF is doing per-destination sharing and won't switch to another EBGP peer unless we reset the EBGP peer with the best route... Is it a correct assumption?


Can 6500 do the per-packet load-share? I don't see this option under interface?


SW003(config-if)#ip load-sharing ?

per-destination Deterministic distribution


SW003(config-if)#


  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
jbrenesj Wed, 08/05/2009 - 10:05
User Badges:
  • Silver, 250 points or more

the Cat6500 does not support per packet load balancing. It will do load

balancing per-dest using CEF on hardware

hardware (PFC) and the Layer 3 switching uses per-flow load balancing based on IP source and destination addresses.

Per-flow load balancing avoids the packet reordering that is necessary with per-packet load balancing. For any given

flow, all PFC equipped switches make exactly the same load-balancing decision,

which can result in nonrandom load balancing.

Thanks for the reply.


Then in this case would CEF switch to another path if the BGP best path never change for the same source and destination sessions?


If the answer is no then "maximum-paths" we have is really not doing anything, correct?


Here are some output from our 6500:


SW003#sh ip cef vrf CRG002 161.178.36.0 detail

161.178.36.0/23, epoch 2, flags rib only nolabel, rib defined all labels, per-destination sharing

local label info: other/9298

recursive via 10.175.202.1

attached to Vlan202

recursive via 10.175.202.2

attached to Vlan202

SW003#

SW003#sh ip bgp vpnv4 vrf CRG002 161.178.36.0

BGP routing table entry for 2:2:161.178.36.0/23, version 2389146

Paths: (2 available, best #1, table CRG002)

Multipath: eBGP

Not advertised to any peer

65309 13979 64591

10.175.202.1 from 10.175.202.1 (10.116.128.120)

Origin incomplete, localpref 100, valid, external, multipath, best

Extended Community: RT:2:2,

mpls labels in/out 9298/nolabel

65309 65000 64591

10.175.202.2 from 10.175.202.2 (10.116.128.121)

Origin incomplete, localpref 100, valid, external, multipath

Extended Community: RT:2:2,

mpls labels in/out 9298/nolabel

SW003#

Edison Ortiz Wed, 08/05/2009 - 11:28
User Badges:
  • Super Bronze, 10000 points or more
  • Hall of Fame,

    Founding Member

BGP, by default, will only install one route in the RIB - the best route.


With 'maximum-path', you are allow to install more than one 'best route'.


The load sharing still relies on CEF which is session based.


In short, by using the 'maximum-path' you are able to share some load via multiple links but the sharing is not 50-50.



__


Edison.

Seems our case is 99-1 load-sharing and one of the remote circuits is saturated.


I have done 20 traceroute from hub to remote and it is always taking the same path, how can I prove to the customer that load-sharing is really working?


I have show them the BGP table, RIB and CEF detail but the fact was remote site is not utilizing both circuits for inbound traffic at all???

Actions

This Discussion