Proxmox VE 3.1 slow NFS reads

[Warod]

Hi,

This issue is directly related to the kernel of your Proxmox host environment. 2.6.32-20-pve and 2.6.32-22-pve does work properly with NFS workloads but newer kernels do not. This affects the NFS mounts done directly by guest OSes regardless the kernel version of the guest. We've tested with multiple 3.x kernels on guests.

Currently running 2.6.32-20-pve on our hosts and there are no issues.

I hope someone else has resources to investigate this issue further and dig up the root causes.

 

[deludi]

Hi Warod,

thank you for the input. I now have better NFS performance by changing our storage platform from Freenas to Omnios+Napp-it.
The throughput has changed from around 10MB/s to around 70MB/s.
Unfortunately i do have an issue: when i backup large vm's (~100GB) the backup freezes, the proxmox node crashes and the vm's on that node are failed over to the other nodes. I managed to work around the backup issue by using ZFS replication (build-in Napp-it) for backups.
The cluster works ok only when i start a vzdump on a large vm i get the above problem.
I am quite sure that this is a kernel issue.

Regards,

Dirk

 

[deludi]

Hello all,

i have an addition to this thread: the freeze problem was not a Proxmox problem but a Linux network driver problem with the Supermicro X9SCM-F mainboards that we used. We have tested with Debian plus KVM and Centos plus KVM: all had the same problem. We have replaced the hypervisors with low power A1SAi-2750F - Supermicro boards. They were rock solid from day 1. The X9SCM-F boards will be used for storage boxes because Omnios+Napp-it has no driver issues with these boards.

Regards,

Dirk Adamsky