[ewg] Problem with NFS/RDMA on CentOS 7 with OFED-MIC-3.5-2-rc1

[Steffen Müthing]

Hi,

I’m trying to get the new 3.3 MPSS stack to work together with OFED-MIC-3.5-2-rc1 on CentOS 7,
but I’ve run into a problem with NFS/RDMA: While I can mount file systems and traverse them,
I cannot access any files larger than about 800 bytes. Trying to cat such a file results in errors like
this:

cat: gcc.lua: Input/output error

or, when copying:

[root at node01 files]# cp test.txt /tmp
cp: error reading ‘test.txt’: Input/output error
cp: failed to extend ‘/tmp/test.txt’: Input/output error
[root at node01 files]# 

At the same time, trying to write a file that is larger than 800 bytes to a server running CentOS 7
tends to freeze that server, while writing to a CentOS 6.5 server works just fine (I just can’t access
the data afterwards).

On the other hand, exporting an NFS share on a CentOS 7 server and mounting it on a CentOS 6.5
client works without any problem.

We are using Mellanox ConnectX-3 VPI adapters (mlx_4 driver) and are currently running kernel version
3.10.0-123.4.4.el7.x86_64. Does someone have an idea of what might be going wrong here?

Thank you!

Steffen Müthing

-------------------------------------------------------------------

Steffen Müthing
Interdisciplinary Center for Scientific Computing
Heidelberg University

Im Neuenheimer Feld 368
69120 Heidelberg
Germany

+49 6221 54 8888
steffen.muething at iwr.uni-heidelberg.de