We do run 2.6 on quite a few clients and 2.5.3 servers in production.
We did this on a set of clients initially for noticeable performance
improvements for some workloads. But it also has helped a bad problem
we've had with LU-5150 causing client crashes, so our 2.6 client use has
grown.
We are running a patched 2.5.3 on the server to try to deal with that
bug, but I don't think it really solved it. The 2.6 client works around
it and seems stable for us.
We are running a patched 2.5.3 server version with non Cray clients running
2.6. We plan to move to 2.7 very shortly. Are Cray clients are 2.5 with the
CLIO 2.6 enhancements backported. For the most part everything works
smoothly. The only issue we have seen since deployment is some read
performance problems with very large block sizes. This is reported under
LU-6370.
On 3/13/2015 8:49 AM, Bob Ball wrote:
Hi,
We are about to set up a new Lustre system here, and we are trying to
decide between using 2.5.3 or 2.6.0 . Files in our older 2.1.6 system
will simply be migrated over, and the old system tossed down the tubes.
Is anyone using 2.6.0 in production? I found this on some 2.6.0
performance data, and it looks rather discouraging.
https://jira.hpdd.intel.com/browse/LU-4841?filter=10828
I have also noticed that the 2.5.3 server rpms actually use a slightly
newer kernel than the 2.6.0 set. By and large, we will use zfs
underlying the OST.
So, I will gladly accept any advice on which way we should turn as we
set this up.
Thanks,
bob
_______________________________________________
HPDD-discuss mailing list
HPDD-discuss(a)lists.01.org
https://lists.01.org/mailman/listinfo/hpdd-discuss
_______________________________________________
HPDD-discuss mailing list
HPDD-discuss(a)lists.01.org
https://lists.01.org/mailman/listinfo/hpdd-discuss