Poor write speeds on new server.

[mwc]

Hi

I've just upgraded from my old HP N40L with 8Gb ram to an Asrock C2750D4I with 32Gb ram.
On the old system I used to have writespeeds of 75-80 MB/s but now on the new system I only get 68MB/s.
I'm using the same disks as before, WD red 3x3TB raidZ. The only disk I've changed is from a 60GB Intel 330 SSD to a 120GB Intel 510 SSD. Other than that my setup is pretty much the same as it used to be.
The only thing I can think of is that maybe one of the 3TB didn't like the move and broke.
Anyone that got some suggestions as to what could have gone wrong or anything to tweak and see if the performance changes?
I should mention that when I transfer a file it start at about 100MB/s and slowly creeps down to 68MB/s (this is when moving files from one folder to another on the same share). Also, some times I've seen it freeze for a couple of seconds.
Would be grateful if anyone knew a fix for this.

Cheers
//Martin.

 

[jgreco]

What's the purpose of the SSD?

 

[mwc]

What's the purpose of the SSD?

I added it as L2ARC cache.

 

[jgreco]

Not likely to be helping, more likely to be hurting by robbing you of ARC.

I would kinda expect the C2750 to be able to push somewhat more than the N40L, but if you're looking at something singlethreaded like CIFS, it gets more complicated. If the pool is more full now than when you established that prior baseline, then this could be completely normal.

 

[mwc]

Not likely to be helping, more likely to be hurting by robbing you of ARC.

I would kinda expect the C2750 to be able to push somewhat more than the N40L, but if you're looking at something singlethreaded like CIFS, it gets more complicated. If the pool is more full now than when you established that prior baseline, then this could be completely normal.

Okay. So you would recommend me to remove the cache then?
Is NFS multithreaded? I run Ubuntu so I can use either.

 

[jgreco]

Unless your pool is so busy that it is unable to expeditiously service multiple requests concurrently, L2ARC is generally a bad idea.

NFS is not singlethreaded in the way that CIFS is. You will probably get different (guessing: better, no promises) results. If you have some patience and take the time to do some tuning, I'd expect the NFS to perform better.

 

[mwc]

Unless your pool is so busy that it is unable to expeditiously service multiple requests concurrently, L2ARC is generally a bad idea.

NFS is not singlethreaded in the way that CIFS is. You will probably get different (guessing: better, no promises) results. If you have some patience and take the time to do some tuning, I'd expect the NFS to perform better.

Okay, is it posible to unmount the cache and keep the raidz intact or do I have create a new pool without the cache?

Guess I'll try NFS instead then. Anything specific that you thought of when you said "tuning"?

 

[jgreco]

Storage->Volumes->View Volumes
Click on ${yourvol} at the top level. That's the one that hopefully says HEALTHY.
At the bottom, a button appears, "Volume Status". Click it.

You'll see "cache", "stripe", "${yourssd}". Click it. There'll be a button appearing at the bottom, "Remove".

Don't do that final step unless you understand what you're doing. It'll remove the device from the pool. For a cache device, that is not problematic.

 

[jgreco]

Guess I'll try NFS instead then. Anything specific that you thought of when you said "tuning"?

Oh, forgot this. Larger buffer sizes? There's lots of good Google bait out on the 'net for tuning NFS, both on the client and server sides.

 

[mwc]

Oh, forgot this. Larger buffer sizes? There's lots of good Google bait out on the 'net for tuning NFS, both on the client and server sides.

Great. Will try it out as soon as I get home. Thanks for all the help.

 

[mwc]

Oh, forgot this. Larger buffer sizes? There's lots of good Google bait out on the 'net for tuning NFS, both on the client and server sides.

Tried NFS now with rsize=8192,wsize=8192 and got about 90MB/s. Seems like 1 disk might be broken though, it kinda stalls from time to time.