Large number of users on freenas 9.2.1.8

Status
Not open for further replies.

cougarmaster

Dabbler
Joined
Jul 7, 2014
Messages
18
Hi all,

I have been using the following system for a few months :-

1] Intel(R) Xeon(R) CPU E5-2620 v2 @ 2.10GHz x 1
2] 16GB Kingston ECC RAM x 8 (128GB total)
3] 4TB WD Enterprise x 36
Created 10 x hdd in RAIDZ2 x 3
6 left over drives are not being used at the moment.
4] Dual port copper 10Gb lagg to a core switch which is confired for lagg
5] LSI 9211 converted to IT mode

About 90 computers are accessing this server. They are mostly using Photoshop, Maya, After effects and Flash mainly others are office documents mainly. I have external AD, AFP and cifs running. Had SCRUB and SMART running but when users were complaining it was slow I turned it all off and seemed to be ok. Also there is a high load and sometimes the drives were reading at high speeds even though the users were not doing much. Is there any thing I missed? Please help really banging my head. I dare not use Autotune as it might do more damage than good. Also the memory it is using has about 20GB left over. Before it would leave about 4-6GB. I cannot have anymore budget to get anything else, what can I do to improve the throughput?

TIA
Eric
 

anodos

Sambassador
iXsystems
Joined
Mar 6, 2014
Messages
9,553
Hi all,

I have been using the following system for a few months :-

1] Intel(R) Xeon(R) CPU E5-2620 v2 @ 2.10GHz x 1
2] 16GB Kingston ECC RAM x 8 (128GB total)
3] 4TB WD Enterprise x 36
Created 10 x hdd in RAIDZ2 x 3
6 left over drives are not being used at the moment.
4] Dual port copper 10Gb lagg to a core switch which is confired for lagg
5] LSI 9211 converted to IT mode

About 90 computers are accessing this server. They are mostly using Photoshop, Maya, After effects and Flash mainly others are office documents mainly. I have external AD, AFP and cifs running. Had SCRUB and SMART running but when users were complaining it was slow I turned it all off and seemed to be ok. Also there is a high load and sometimes the drives were reading at high speeds even though the users were not doing much. Is there any thing I missed? Please help really banging my head. I dare not use Autotune as it might do more damage than good. Also the memory it is using has about 20GB left over. Before it would leave about 4-6GB. I cannot have anymore budget to get anything else, what can I do to improve the throughput?

TIA
Eric
Scrubbing is rather resource intensive. You should do it, but not necessarily more than twice per month. Schedule them for when you have less users. You should also be running SMART. Do not touch autotune.

Tracking down performance problems is HARD. Heck, even benchmarking is hard. Here are a few random thoughts:

Your pool only consists of 3 vdevs. Each RAIDZ can have the IOPS performance of a single disk. You can increase performance by increasing number of vdevs. For instance, you have 36 drives. You can try 6 * 6-disk RAIDZ2 vdevs or even switching to mirrors.

The drives thrashing when users aren't doing much can indicate that samba's trying to mess with DOS attributes. For instance, when I open \\server\foo and run "dtruss -p <pid of smbd session>" I can see samba schizophrenically reading DOS attributes from all over the share. This is a very inefficient process. You can try disabling DOS attributes per steps here:
https://forums.freenas.org/index.ph...-to-improve-samba-browsing-performance.24906/
It does carry some caveats. If this alleviates your performance problems you'll want to remove "write attributes" permission from your users.
 

cougarmaster

Dabbler
Joined
Jul 7, 2014
Messages
18
Hi anodos,

Thanks for the reply. Also the DOS attributes is not a way for me to go. I guess I am stuck until I can reconfigure all the vdevs. So what gives on the arc not using all the ram? Is it because of a problem on the system or it is working? Would there be any great improvements if I add SSD cache and zil? Also what would be the best way to calculate my SSD size if I am to get them for cache and zil? Most of the files would be in the regions of about 10 - 20MB in size for the images and 1-2MB for other document files. Also all the Maya and After Effects are rendering back to this server as it is reading from it.

TIA and appreciate the time
Eric
 

anodos

Sambassador
iXsystems
Joined
Mar 6, 2014
Messages
9,553
You can use arc_summary.py to view stats regarding arc as well as current arc settings (which should give insight into that 20GB).

It is possible that you'll benefit from an L2ARC device. With 128GB you're in territory where it may make sense to have one. L2ARC takes time to 'warm up'.

There are some tests you can do to see whether you'll benefit from a ZIL. Just search around the forums in the stickies.

If you have an AD domain you can push out a GPO update to disable thumbnail generation. If you have any XP machines, get rid of them. SMB2 is much more efficient.

Otherwise you're probably stuck with adding more vdevs either by restructuring your pool or putting in a shelf with lots more spindles.

I'm not a zfs guru and so I can't do much more keep things relatively vague. If you need more specifics you can ask on IRC or contact iX for a one-off support session. Sometimes paid support can help you avoid making expensive unnecessary hardware purchases.
 
Last edited:
Status
Not open for further replies.
Top