[one-users] NFS datastore performance during VM instantiation

[cloudprovider]

Hi, Matt


I have two solutions for this issue, for reference only:


1. Try qcow2 snapshot for creating new VMs in KVM environment
2. NFSv4 might be another option with server-side copy feature


Both could avoid coping large files between NFS server and client.


Thanks,
Lei




------------------ Original ------------------
From:  "Matt Willsher"<matt.willsher at boxuk.com>;
Date:  Fri, Jun 22, 2012 07:06 PM
To:  "users at lists.opennebula.org"<users at lists.opennebula.org>; 

Subject:  Re: [one-users] NFS datastore performance during VM instantiation



  
Hi Ruben,
 
 
 
Thanks for your reply. We came to the same conclusion in the end; we need a more suitable file system to store the data. I did try to tune the NFS client without  any improvement and the NFS server (an EMC VNXe3100) doesn’t have anything to tune, though it does use async out of the box.  We’ll investigate other options.
 
 
 
Best regards,
 
Matt
 
 
 
 
 
From: Ruben S. Montero [mailto:rsmontero at opennebula.org] 
 Sent: 22 June 2012 11:41
 To: Matt Willsher
 Cc: users at lists.opennebula.org
 Subject: Re: [one-users] NFS datastore performance during VM instantiation
 
 
 
Hi Matt,
  
 
 
  
NFS is not a valid solution for medium size deployments, specially when there is contention on the network (deploying multiple VMs at the same time, multiple VMs with heavy I/O concurrently). The solution could be based on two basic alternatives  (or a combination of both):
 
  
 
 
  
1.- Use a better NFS server (SAN) or a better distributed/shared FS (gluster,  lustre...)
 
  
 
 
  
2.- Balance the NFS load by using different servers. Simple configure several datastores mounting volumes from different NFS servers. You may also want to tune the NFS mounting parameters caches, syncs,.....
 
  
 
 
  
Hope it helps 
 
  
 
 
  
Ruben
 
  
 
  
On Fri, Jun 15, 2012 at 12:43 PM, Matt Willsher <matt.willsher at boxuk.com> wrote:
   
Hi,
 
 
 
I’m currently working on Open Nebula using NFS as a shared data store. I’ve found that during the large copy done during instantiation all other VMs on the same node suffer storage  performance issues and are unable to write to disk. In some circumstances this produces warnings in the VM’s kernel due to time outs. I’ve been over the system configuration, tried different NFS options, sync, async, and others. Nfsiostat is showing avg exe  times to be 40+ seconds during these large sequential writes.  There are no errors in either NFS or network stats and the NFS server is responsive when accessed from other clients. 
 
 
 
My question is, has anyone else come across these sorts of performance issues  during VM instantiation when using NFS a shared datastore? 
 
 
 
Thanks,
 
Matt 
 
 
 
 

 
 
 

 _______________________________________________
 Users mailing list
 Users at lists..opennebula.org
 http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
 
 

 
 
  
 
 
 
-- 
 Ruben S. Montero, PhD
 Project co-Lead and Chief Architect
 OpenNebula - The Open Source Solution for Data Center Virtualization
 www.OpenNebula.org |  rsmontero at opennebula.org | @OpenNebula
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opennebula.org/pipermail/users-opennebula.org/attachments/20120623/d6ede8b5/attachment-0003.htm>