CIFS Performance Benchmarking

we have been using IOmeter. Good tool and free

I've used iometer some time ago. I can run the workers on test workstation?

The trouble with IOMeter is that it doesnt reflect the load that a normal Window file server sees

If you buy a NAS server to attach one or a few clients and mostly care about MBytes/sec then IOMeter is fine

Normally though you buy a Celerra to serve hundreds or thousands of Windows clients simultanously.
These clients will not just do read/write but also a lot of metadata ops and stress the network stack as much as raw disk performance

I know these single client "benchmarks" are convinient to run - but its questionable what value their results have if they dont reflect your workload

Unfortunatley the only benchmarks with multi-client workloads closer to a Windows file sharing environment arent free.

That's exactly where I'm struggling. Which specification most resembles file server access. For my purposes I need only show that the Celerra as faster or faster than the current windows file server. To that end as long as my test is consistent I think I can make a reasonable interpretation.

sure ..IOmeter will not represent true CIFS environment you have out there but if i compare apples to apples it gives me a pretty good idea. I take Celerra share ..beat it up with IOmeter configured with a few workers, i set my workload profile to 80% read/20% write , random IO ( i believe this represents well my user's access pattern). Once done ..i do the same with a windows file server share.

We did a very similar thing several years ago comparing CIFS performance between NetApp and EMC. First we looked at the environment in question and gathered detailed CIFS operations statistics. Then we found a tool (I think it was called CIFS torture, later renamed to CIFSBM) We tried to duplicate the workload as best we could and then pounded both platforms to see the performance results using the created workloads.

Keep in mind that CIFS has lots of different operations that have very different resources usage. As long as you use the same on both platforms you should be fine, but I would try to duplicate the production traffic as much as possible.

thanks sagle ...where can i find this tool ? I've google a little but can't find anything.

Try searching for "cifs-load-gen" or "cifs_bm". It is a linux tool that we used several years ago. It may have morphed or changed since then. I have the performance numbers from our testing, but it was comparing a NetApp 980 to a NSX xblade 60. I don't have any numbers for windows platforms.