[Lustre-discuss] lctl tcp routing

[Mohr Jr, Richard Frank (Rick Mohr)]

Are you using the same LNET names for the two file systems (ie - tcp0)?  Do the compute nodes have a single ethernet interface for both file systems, or do they have two (one on 172.16/16 and one on 192.168.200/24)?

-- 
Rick Mohr
Senior HPC System Administrator
National Institute for Computational Sciences
http://www.nics.tennessee.edu


On Nov 20, 2014, at 9:15 AM, Alexander Oltu <Alexander.Oltu at uni.no>
 wrote:

> Hi,
> 
> We have  a cluster with its internal Lustre filesystem with 10GbE
> TCP/IP lnet. The IP range is in 172.16/16 subnet.
> 
> Now we want to mount an external Lustre filesystem on our compute
> nodes. The external FS has an IP range in 192.168.200/24 subnet.
> 
> Switches from both are connected.
> 
> We fixed routing tables on both subnets by:
> on 172.16 machines:
> route add -net 192.168.200.0/24 dev eth0
> on 192.168.200 machines:
> route add -net 172.16.0.0/16 dev eth0
> 
> We can ping nodes each other from the both subnets. But lnet ping is not
> working and as such we can't mount external Lustre filesystem.
> 
> I know that we can add virtual IPs on the compute nodes in 192.168.200
> range and it will work, but maybe someone knows about solution with
> routing tables and lctl routes?
> 
> Thanks,
> Alex.
> 
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss