We have a problem in load-balancing three point-to-point links between
routers called Kinnegad and Heuston. The links are DCE-DTE
between serial interfaces, each clocked at 2,000,000 bits/sec and
configured similarly. They land on Kinnegad:Serial1/1, 1/2 and 1/3.
There's a BGP session between the routers, each knowing the
other by its loopback 0 address. At interface level, OSPF
is running on the three links, by including 193.1.195.0/24
(the supernet of the three /30 link network addresses) as
a network in area 0 at both ends. This exploits OSPF's
load-balancing capabilities, with the aim of roughly equating
traffic across the three links.
In practice, Serials 1/1 and 1/3 are well balanced, but less
traffic goes across 1/2. The result is that the total throughput
is limited to just over 4Mbps, instead of 6Mbps. The reason
is that OSPF sees routes via the three links for some networks
only; for the majority, it sees routes only over Serials 1/1 and
1/3.
On examination, the three-link routes are seen to apply to OSPF
intra-area (IA) routes only. These routes belong to live interfaces
explicitly declared to be in area 0 by an OSPF process somewhere
in our WAN. Routes seen over Serials 1/1 and 1/3 only have been
redistributed into OSPF from RIP or static and are seen as OSPF
external type 2 (E2).
The first question is why does OSPF on Cisco manifest the distinction
between IA and E2 routes like this, advertising the latter over only
two of the three parallel links.
Second, how do we fix this centrally, without having to re-visit OSPF
processes at each site, let alone forcing all routes to originate
in OSPF area 0?
Any help much appreciated.
Mike Norris
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:18 EDT