Measure NFS Throughput

[Greg Stark_1]

We have an rp2470 hooked directly to a NAS box via copper gigabit. I would like to the measure the throughput of this connection. So far, I have created a 2GB file on one of the 2470's nfs mount points and copied it to another nfs mount point:

time cp /nfs1/file.tst /nfs2/file.tst

This test usually takes about 105 seconds which is about 20 MB/s each way, or total of 40 MB/s.

I was just wondering if this is the best way to test this configuration. Is there a lot of overhead with this test and I'm not getting true results?

Thanks again,
Greg

[Steven E. Protter · Accepted solution]

Try nfsstat

from the manual

-m Display statistics for each NFS mounted file system. This
includes the server name and address, mount flags, current read
and write sizes, the retransmission count, and the timers used
for dynamic retransmission. The srtt value contains the smoothed
round trip time, the dev value contains the estimated deviation,
and the cur value is the current backed-off retransmission value.

-n Display NFS information. NFS information for both the client and
server side will be printed. Can be combined with the -c and -s
options to print client or server NFS information only.

-r Display RPC information.

-s Display server information.

-z Zero (reinitialize) statistics. This option is for use by the
super-user only, and can be combined with any of the above
options to zero particular sets of statistics after printing
them.


For other collection of info in the background, I'm attaching a tested, production, script that you can control the time of data collection which occurs in the background.

SEP

[Leif Halvarsson_2]

Hi,
It depends what you are interested in, it can give an idea about raw transfer rate.
Another useful test is the PostMark benchmark, you can download it from NetWork Appliance:
http://www.netapp.com/tech_library/3022.html

[Mark Landin]

This will tell you your throughput. What it won't tell you is what part of the chain is the bottleneck (is it network latency? file system speed? mis-configured NFS daemons or settings? retransmissions?). Without that information, you don't know what to tweak to make an improvement.

Check this excellent paper on NFS:
http://h21007.www2.hp.com/dspp/tech/tech_TechDocumentDetailPage_IDX/1,1701,952,00.html

It may help dig deeper into the mysteries that are NFS.

[Stefan Farrelly]

There is an awful lot to consider for nfs NAS performance. We use Netapps devices and its takend us some time to get the best performance (and elimintate nfs errors). We found this doc very important;

http://www.netapp.com/tech_library/3146.html

Just for its extensive coverage of kernel parameters, patches, nfs mount options, ways to test throughput etc. Have a read.

We also found the excellent book "Optimizing nfs performance" (on amazon.com) very helpful. Its lists hidden kernel parameters for nfs, and many many other things to look into and consider. It also lists things not to do. Well worth the money.

[Mark Landin]

Stefan, if that is the book by Dave Olker, he's the authoer of the paper I cited in my earlier response. I thought his paper was very thorough ... I can't imagine what he did with the 200 extra pages that went into the book. :)