NAS testing

[Jtyler]

Maybe someone can advise on NAS testing issue. . .When using IOzone to run NAS tests for comparison with NAS charts (64k record size, 32M - 1G file size) the results are not believable for file sizes smaller than available RAM. Basically the results show transfer rates much faster than the drive is capable of. When the file size reaches the RAM size (512MB) there is a noticeable drop off, and the results are believable. Is something being done with the test to negate the caching/buffering effects?

The only difference from the "How We Test: Networked Storage Devices" article and my system is I'm running IOzone on a Linux box instead of windows and I'm using NFS.

Just trying to get an apples to apples comparison.

 

[Madwand]

I agree -- I don't find "small file size" tests to be meaningful, and typically only consider the large file sizes -- 1 gig or so.

It's not a Linux vs. Windows issue, as I've seen the same behavior under Windows.

 

[thiggins]

The highlighted note at the top of each NAS Chart explains this. Whenever you see performance that is higher than the network connection supports, you are seeing cached performance.

Looking at read performance provides a better idea of actual "hardware" performance.

Even though it is cached performance, the small filesize plots represent real performance. Caching is why many people say they see performance higher than shown in the NAS charts when they do timed transfer tests.

Some people like the small filesize tests, which is why they were added.

 

[Jtyler]

Here's my issue, lets use the Synology DS508 as an example. The 1000Mbps read performance chart shows throughput around 54MB/S with file sizes between 32MB and 256MB, then the speed drops off when the file size reaches the RAM size. All the results are very believable.

Trying to get results from my system for comparison, I see read speeds around 100MB/s with file sizes between 32MB and 512MB, then the throughput drops off. The results after the drop off are believable, but 100MB/s sustained read speeds from the drive is just not reasonable, therefore I wonder what is being done to mitigate this effect in the Synology chart and all the other charts for that matter.

 

[thiggins]

Are you running iozone on the machine that you are trying to benchmark, or on another networked machine? It should be the latter.

 

[Jtyler]

Its running on another networked machine.

 

[thiggins]

What is the configuration of the machine that you're testing, the machine you are running iozone on and the network connection?