I'm not an expert on this, but here is what I would try:
1. Are you using the latest virtio drivers on your Windows guest(s)?
http://alt.fedoraproject.org/pub/alt/virtio-win/latest/images/
2. Is there any particular reason you use CentOS 6.4 on your storage server?
I would update it to CentOS 6.5 so you would have matching Linux
kernels on both of your systems.
Zoltan
On 2/19/2014 11:21 AM, engineer at colocat.ru wrote:
> Hello. We've started a virtualisation project and got stuck in one moment.
> Currently we are using the following:
> Intel 2312WPQJR as a node
> Intel R2312GL4GS as a storage with Intel Infiniband 2 ports controller
> Infiniband Mellanox SwitchX IS5023 for commutation.
>
> The nodes run CentOS 6.5 with built-in Infiniband package (Linux v0002
> 2.6.32-431.el6.x86_64), the storage - CentOS 6.4 also built-in drivers
> (Linux stor1.colocat.ru 2.6.32-279.el6.x86_64).
>
> On the storage is made an array, that is shown in system as /storage/s01.
> Then it is exported via NFS. The nodes connect to NFS by:
> /bin/mount -t nfs -o
> rdma,port=20049,rw,hard,timeo=600,retrans=5,async,nfsvers=3,intr
> 192.168.1.1:/storage/s01 /home/storage/sata/01
> mount shows:
> 192.168.1.1:/storage/s01 on /home/storage/sata/01 type nfs
> (rw,rdma,port=20049,hard,timeo=600,retrans=5,nfsvers=3,intr,addr=192.168.1.1)
>
> Then we create a virtual machine with virsh with a disk bus virtio.
> All is OK, until we don't start Windows on KVM. It may work for 2 hours or
> 2 days, but under heavy load it hangs the mount (i.e. /sata/02 and 03 are
> accessible, but requesting 01 will result in a total hang of console).
> This can be beaten only by hardware reset of the node.
> If we mount without rdma - all is fine.
>
> What can we do in such case? If any debugs or info needed - please ask.
>
> Best regards, Nikolay.
>
> _______________________________________________
> CentOS-virt mailing list
> CentOS-virt at centos.org
> http://lists.centos.org/mailman/listinfo/centos-virt
>