[Lustre-discuss] testing hw for lustre--mounting as client on ost

[Ms. Megan Larko]

Hello,

In trying to create a lustre filesystem on some new hardware I did
some benchmarks on the hw as ext3.
Now I am trying to benchmark with the same bonnie parameters under
lustre.   I am trying to keep the test on the new hardware and off of
the live system with the exception of the mgs which I am using with a
new disk as mdt to benchmark.

I have found that I cannot mount the disks as a client on the oss to
which those disks are mounted/connected.   I get a timeout Lustre
Error 11-0 and many processes of
[root at oss4 ~]# ps -ef | grep ost
root     12880     1  0 12:25 ?        00:00:00 [ll_ost_00]
root     12881     1  0 12:25 ?        00:00:00 [ll_ost_01]
root     12882     1  0 12:25 ?        00:00:00 [ll_ost_02]
root     12883     1  0 12:25 ?        00:00:00 [ll_ost_03]

Andreas response to the same error of another poster in Sep 2007 said
that high load can prevent the oss from mounting its own disks as a
client.  I have no load on this new, non-live oss other than the small
test volumes I have set-up.    There is no other load.   Can it not be
done because the ip-traffic (on o2ib in my case) just conflicts?

megan


******************   Andreas original response to the error
***********************
On Sep 03, 2007  20:13 +0300, vorl wrote:
> Sep 3 19:44:58 snapxtc3 kernel: Lustre:
> 3631:0:(ldlm_lib.c:724:target_handle_connect()) snapxtfs-OST0000: refuse
> reconnection from d9f5edc9-a349-4483-5c2f-7fdeaae46d60 at 0@lo to
> 0xffff81012963d000/3

Running clients and servers on the same node can deadlock under high load.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

******************   Andreas original response to the error
***********************