Poor RAID-6 performance with Adaptec 7-series

[Alexdi]

I have this 6 x 6TB RAID-6 with this configuration:

Windows 10 Pro 1803
Intel X3450
Supermicro X8SIL-F
2 x 8GB DDR3 ECC
Adaptec ASR-78165 RAID HBA
(White label) WD Red Pro 6 TB 7200 RPM x 5
(White label) Hitachi He6 6TB 7200 RPM x 1

It's 72% full. This is what I'm seeing:



CPU was at 5% while this test was active. 11GB of free RAM.

RAID configuration:



This seems pretty abysmal to me. I just expanded the array with the Hitachi. The homogeneous WD 5 x 6TB configuration (~95% full) was 30 MB/s slower on both sequential operations and 1.X on the other fields.

Is this typical? If not, what am I missing?

 

[Myth]

Run the speed test on the drive itself without going through adaptec. I assume these are SATA? connect directly to the MB via a SATA cable and see what kind of speeds you get on each drive individually.

Also what is connector 0 and 1, why do you have two different connectors with different drives under different connectors?

Also you can try to run mydefrag a free defragmentation program which will work great to improve performance. Run data disk weekly if you don't have an OS on that raid array.

Also if you only have 6 drives maybe RAID 5 is better?

 

[moblaw]

Same raid-controller, different drive-configuration.

15x6 TB mixed vendors, (hgst ultrastar, Seagate, wd gold.)

​

6x8 Hgst HE8 SAS

​

pooled together, with ssd write-cache (drivepool config).

​

I've been using this for 2 or more years, contemplating zfs every now and then, it's our only way of getting higher IOPS, with the same 3,5" configuration. Just add some nvme/slog/arc drive and away you go. but the easy management of Adaptec has me hanging on, for now.
dont use raid5. Are you 100% on sata3? Whats your score without the onboard raid-cache/disabled. And are you also using the newest 71865 firmware?

can you test with atto benchmark? It gives a better overall image.

 

[Myth]

We use 81605zq adaptec controller cards with 16x Seagate SAS 12 drives 6TB each. We get well over 3000MBPS r/w.

RAID 6

 

[j_h_o]

When I had perf problems previously, the card was either overheating (is it getting the airflow it needs?) and/or there was issue mixing SAS/SATA drives with the backplane in the chassis.

 

[Alexdi]

Run the speed test on the drive itself without going through adaptec. I assume these are SATA? connect directly to the MB via a SATA cable and see what kind of speeds you get on each drive individually.

Also what is connector 0 and 1, why do you have two different connectors with different drives under different connectors?

Also you can try to run mydefrag a free defragmentation program which will work great to improve performance. Run data disk weekly if you don't have an OS on that raid array.

Also if you only have 6 drives maybe RAID 5 is better?

SATA drives, yes. The controller has 4 breakout cables, each with 4 ports. Each cable is a connector. I don't think I can test drive speeds on the board directly without risking a rebuild, I've no reason to believe the drives are atypical. Not into RAID-5, rebuild times on this array are about 20 hours. Too much risk. It turns out the array is 14% fragmented, so I started a defrag after benchmarking. I didn't realize that was necessary with Windows 10, thanks.

Same raid-controller, different drive-configuration.

15x6 TB mixed vendors, (hgst ultrastar, Seagate, wd gold.)

View attachment 8629​

6x8 Hgst HE8 SAS

View attachment 8630​

pooled together, with ssd write-cache (drivepool config).

View attachment 8631​

I've been using this for 2 or more years, contemplating zfs every now and then, it's our only way of getting higher IOPS, with the same 3,5" configuration. Just add some nvme/slog/arc drive and away you go. but the easy management of Adaptec has me hanging on, for now.
dont use raid5. Are you 100% on sata3? Whats your score without the onboard raid-cache/disabled. And are you also using the newest 71865 firmware?

can you test with atto benchmark? It gives a better overall image.

Caches off:



Caches on:



Read cache off:



It's faster with the read cache off? I don't know what's going on. Less bad than before, though.

 

[Myth]

That's interesting. You might consider a replacement adaptec card? If you get the same kind it should import with no issue. But can't be sure.

As far as disk defragmenter goes, I find the Mydefrag tool beats windows defrag tool by a performance ratio of 10 to 1. So I'm not clear if you are using the windows defrag tool or the MyDefrag tool. But yes we see huge performance gains by using mydefrag tool on our large arrays.

 

[jerryxlol]

hmm first what i see is absence of pcie3.0, i got result with my ada 78165 with cache on. (Stripe 1024, IPM disabled ). Drive is full 90%. and controller has two raids 5 (4x6T) and 6(8x3T). Crystalmark says something similar to you.

Second photo just lenght rised to 2GB (to oversize the cache). Idk what are you using this setup for, but for me it is just NAS

 

[Myth]

those 6,000Mbps marks are just reading the controller cache. You should try the AJA speed test tool. I find it cuts through that cache and you can read the real HDD performance.

*Edit - oh wait, I see now, you just changed it to 2GB to cut through the cache.

 

[Mike W]

Hello,
Your Max cashe is not enabled on this card which i am assuming (Always used LSI) allows the use of ssd as add-on cashe . Your write through is on by default and should only be turned off if you have some other cashe with either battery BU or they actually use a module with a super capicitor that writes any data to a to non volotile memory module if there is loss of power . I know some of the LSI cards will automatically turn write through back on if there is a battery failure or it is not connected to prevent drive array corruption.
Check to ensure you have the super capacitor module or battery BU installed if it not insure that your model card is supported.

Please see link below
http://download.adaptec.com/pdfs/readme/microsemi_series-7-controller_readme_2_2017.pdf