Running iozone processes onboard an Isilon node isn't good practice at all.  The best way to baseline a given client (in your case, NFS) is to directly connect via a dedicated LAN segment w/o any switches/routers in the path.  Isilon itself tests in this fashion to establish baseline performance of nodes.  If you are having less-than-expected performance on NFS, start with the NFS client best practice paper (found on EMC One).  In the vast majority of cases this turns out to be an exercise in NFS client tuning.

Customers don’t have access to EMC One

Sure, but your SE does.

To get the doc from the support site, try KB article 90041

Absolutely there is a specific reason.  You shouldn't be installing any 3rd party packages at all on an Isilon node.  The only packages (.pkg, etc.)you should ever install are those which come from Isilon.

Trying to save you a lot of headache down the road.  There is no support for using a OneFS node as a build platform for 3rd party software, both corporately and in actual practice.  A OneFS node is a product.  Running 'make' to roll your own on a node is, strictly speaking, a violation of your warranty.  You're altering the product outside of agreed-to mechanisms.

Having said all that, again, the best way to ferret out client-side perf problems is to connect that client directly to a LAN segment.  Capture the packets with Wireshark or some other app.  Have a look.  Use the existing best practices that we and the community provide.  There are a ton of smart folks on this reflector.

Rob,

Is there any specific reason why installing one package would be considered "not a good practice"? iozone files don't even leave outside the directory where the source is unzipped?

Just curious.

Thanks.

>   Capture the packets with Wireshark or some other app.

...and not to forget the formidable trio of isi statistics client/heat/drive

before bringing along the sledge hammer

-- Peter

try the isi statistics commands.

Examples:

Overall Cluster Utilization

isi statistics pstat --top

Drives in Chassis

isi statistics drive --top

Busy drives

isi statistics heat

Other Options

'isi statistics' options are:

  --help, -h  Print usage help and exit

'isi statistics' sub-commands are:

  protocol   Protocol mode displays statistics organized by protocol, such as

             NFS3, HTTP, and others.

  client     Client mode organizes statistics by users accessing the cluster.

  system     System mode displays general cluster statistics.  This mode

             displays op rates for all supported protocols, as well as network

             and disk traffic (in kB per second).

  query      Query mode provides highly customizable access to any statistic

             in the cluster statistics library.

  history    History mode provides basic access to historical values of

             statistics which are configured to support history.

  pstat      Pstat mode displays a selection of cluster-wide and protocol data

  heat       Heat mode displays most active /ifs paths for a variety of

             metrics

  list       List valid arguments to given option.

  describe   Print documentation strings for given statistics.

  drive      Drive mode shows performance by drive.