cancel
Showing results for 
Search instead for 
Did you mean: 

NAS/NFS performance query

 
Adam Noble
Super Advisor

NAS/NFS performance query

Hi all,

I have one of those interesting situations in which we have all been in.

We have upgraded one of our services to Itanium based machines running 11.31. They are faster machines, more memory and one significant change the NFS is now being presented from an EMC cellera storage array. It was previosuly presented from other servers in the solution. The network links are the same 1GB. I also have a dedicated link on the NAS purely for this solution.

Basically all the system does all day is get files dumped via ftp to the NAS share. I appreciate NFS is never the most reliable product but I did not expect degredation in performance and the odd error for good measure as rare as they are:-

vmunix: WARNING: Synchronous Page I/O error occurred while paging to/from NFS server

I have a call with EMC but I wondered if anyone here has any ideas. Surely performance should be better to a dedicated NAS array.

2 REPLIES
Steven E. Protter
Exalted Contributor

Re: NAS/NFS performance query

Shalom,

The logs on the EMC server need to be checked.

This error points to a problem on the NFS server, which is according to your post providing NFS services.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Mel Burslan
Honored Contributor

Re: NAS/NFS performance query

Also, since you have changed the architecture, the protocol you are using to access the nfs server might have been changed to the worse (read slower but more robust). Usually, in the early days, NFS ran mostly on UDP protocol. In the days where 100MB Ethernet ports were a luxury, every bit you saved by not acknowledging every packet as required by TCP, made a lot of sense.

Today, you can buy a 8 port gigabit home switch with the pocket change from a retail electronic store. So, the network speed stopped becoming an issue (for the casual user that is) and things started to TCP direction for the transmission integrity it provides.

So, your EMC array might be serving NFS over TCP, whereas your old server was doing it over UDP. Normally, it should not make the whole world difference if your bursts are sporadic and steady data rates are low, but if your organization uses NFS as their primary data transfer method and the data we are talking about is very large chunks, then all bets are off.

Last butr not the least, Celera arrays are not know to be speed daemons. They are pretty much the bottom feeder of the EMC product family. As a test, I can suggest you to get your hands on a fairly modern X86 machine (preferably server class) and a copy of the FreeNAS linux distribution and set it up on your network. If you have a VMWare ESX server somewhere in your arhitecture, I think a variant of FreeNAS is available as a VMWare appliance image somewhere on the Internets. So, run an independen NAS/NFS server ans see if you are running into the same issue.

At least you can be productive while waiting for the EMC behemoth to move.

Good luck.
________________________________
UNIX because I majored in cryptology...