NAS

NAS Performance evaluation procedure (put yours, I'll compile a mixture)

3 1592

Hi !

I'll edit this post later with your input information.

Can you tell me how to evaluate my NAS performance ?

What software to use ?

For which part is this or that software appropriate for ? (HD speed, network speed, overall performance, etc.)

What information can we post on the forum to make troubleshooting or benchmark easy to evaluate ?




.. I just think it would be good for beginners (like me) to know what to check, how to check it, and how to post information so the more knowledgeable people can read their questions more easily.


What I've found/gather so far :

storage performance
iozone
bonnie++



network performance
iperf
netperf
details on how to adjust parameters like record/block/window sizes


Board/overall performance

htop

dstat -cdnpmgs --top-bio --top-cpu --top-mem

Helios' LanTest (usr: tools, pwd: tools) ... post screenshot

Add post (Tue May 5, 2015 06:09):
File transfert with windows explorer / Mac finder / Linux Rsync ... check transfert speed.
All these tests are good to identify issues and potential for optimization. But don't forget real world tests with files and workloads that you acutally use.
Just to give you one example: When I upgraded from the older 3.4 kernels to the mainline kernel 4.0, I ran a few quick tests with Helios LanTest. I found that certain test profiles (I think it was the one for Gigabit LANs) gave horrible values in one direction (something like 3MB/s read speed or so) while the other tests showed the expected values of about 29-30 MB/s (with an USB harddrive attached). Now, that first looked scary, but when I actually transfered files with either Windows Explorer, Finder (on a Mac) or rsync on Linux, I could not reproduce this odd behaviour but saw good readings like 29MB/s for large files and 22MB/s for lots small files instead. At this point I decided to not further dig into it, because it had no practical relevance for me.

So, in addition or before running a series of sophisticated tests, you should test your daily usage and see if you're satisfied with it. This might save you some time.

Edited by chrisprols at Tue May 5, 2015 06:09

well noted.
I'll edit the first post to include a file transfer test from windows/finder/rsync.
thanks for your input.


.... actually I can't edit an old post
I added your input at the end of the post.

tkaiser  
Edited by tkaiser at Tue May 5, 2015 07:00
silentcreek replied at Tue May 5, 2015 06:01
All these tests are good to identify issues and potential for optimization. But don't forget real world tests with files and workloads that you acutally use.


I second that to first have a look for real world performance and only dig deeper into it when problems arise since dealing with this network stuff requires some deeper knowledge to interpret results correctly.

The aforementioned LanTest uses different record sizes depending on the setting chosen (explained here in detail: http://www.helios.de/web/EN/support/TI/157.html). And it might make a huge difference whether chunks of 128 or 1024KB will be sent over the network (especially when latency is also an issue). And the link also explains how Explorer/Finder might be able to saturate a link that's misconfigured horribly (eg. due to 'tcp delayed ack' problems) since they use super large block sizes and asynchronous reads or writes in parallel. The LanTest performance is therefore something you could expect from within applications (eg. opening a large Photoshop document directly from the server in an image editor)

But when you encounter real performance problems then it's always a good idea to test 'from bottom to top' (server's filesystem isolated, network throughput isolated, both combined using LanTest with different settings)

You have to log in before you can reply Login | Sign Up

Points Rules