[Beowulf] how fast can NFS run?

[ctierney at hypermall.net]

Quoting Bruce Allen <ballen at gravity.phys.uwm.edu>:

> I'd like to know the fastest that anyone has seen an NFS server run, over 
> either a 10Gb/s ethernet link or a handful of link aggregated 
> (channel-bonded) Gb/s ethernet lines.
> 
> This would be with a small number of clients making large file sequential 
> reads from the same NFS host/server.  Please assume that the NFS server 
> has 'infinitely fast' disks.
> 
> I am told by one vendor that "NFS can't run faster than 100MB/sec".  I 
> don't understand or believe this.  If the server's local disks can 
> read/write at 300MB/s and the networking can run substantially faster than 
> 100 MB/s, I don't see any constraint to faster operation.  But perhaps 
> someone on this list can provide real-world data (or say why it can't 
> work).

I built some servers a couple of years ago and was able to run
NFS at about 100 MB/s.  The system was Dual socket Xeon, single gigE link, 1
Gb/s fibre channel, XFS filesystem, jumbo frames.  The limiting factor was that
the local filesystem could only do 100 MB/s.  Jumbo frames increased the
performance from 60 MB/s to about 95 MB/s.

I know those servers could have gone faster if I had another gigE link
and faster disk.  Going above 100 MB/s shouldn't be a problem.

If you want some other evidence, go look at www.agami.com.  They are a
relatively new storage applicance company.  They claim their servers can
provide
up to 1000 MB/s over NFS (quad socket opteron, linux+xfs, lots of disks).
I don't know if they tweaked the kernel or NFS to get this performance, but
from the hardware they use I believe they could get that performance.  
It certainly should be able to sustain more than 100 MB/s.

I am not endorsing the product, I am just pointing out some data that may
be able to help you to determine what you can/should get out of a Linux Based
NFS server.

Craig



> 
> Note: I am free to use modern versions of the NFS protocol, jumbo frames, 
> large rsize/wsize, etc.
> 
> Cheers,
>  	Bruce
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
> 




----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.