[Xorp-users] OSPF point-to-multipoint issue

[Ben Greear]

On 06/18/2010 09:03 AM, Vincent, Michael - 0665 - MITLL wrote:
> We're doing some testing with 6 Redhat VM's.
>
> # uname -a
> Linux nlet-89-1 2.6.18-194.el5.g63 #1 SMP Wed Apr 7 00:08:09 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
>
> All the VM's have an interface to 172.16.1.0 /24 on which OSPF is configured in point to multipoint mode creating a full-mesh.
>
> We're using a linux kernel module to selectively drop packets between certain hosts (based on MAC address) to simulate link loss.  We expect OSPF to route around the failures by multi-hopping to hosts it can still reach.  This works fine.
>
> ----
> EXAMPLE:
> - host IP's are host numbers
> - IP's are on interface eth1
> - eth1 is OSPF point-to-multipoint *
>
>          host1      host2
>
> host3     172.16.1.0/24    host4
>
>      host5              host6
>
> * Example config for host1
> protocols {
>      ospf4 {
>          router-id: 10.10.1.1
>          area 0.0.0.0 {
>              area-type: "normal"
>              interface eth1 {
>                  link-type: "p2m"
>                  vif eth1 {
>                      address 172.16.1.1 {
>                          hello-interval: 1
>                          router-dead-interval: 4
>                          neighbor 172.16.1.2 {
>                              router-id: 10.10.1.2
>                          }
> ...
> ----
>
> As we simulate loss of link between host1 and host3, host1 continues to forward packets to host3 via a new host route created to host5.  This is verified by a "route -n" from the linux command line also.
>
> ISSUE:
> When we re-establish the link between host1 and host3, host1 *CONTINUES* to forward packets to host3 via the host route created to host5.  OSPF does *NOT* recover and remove the host route from the table.  This is verified by a "route -n" from the linux command line also.

If you look at the OSPF neighbors, are they properly re-established when you rejoin the link?

We do the same test, except with emulated ethernet links instead of 'p2m' connections,
and the rejoins appear to work correctly.

Are you using ospf costs to make one route more preferred than the other?  If all costs
are the same, then it won't flip back to the original route.

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com