Wildly Varying Parity Check Speed. CPU, disks, controller don't appear to be bottlenecks....flaky disk?

[RealActorRob]

I do one a week, and there's one going on right now due to a hard reset. Knock on wood, zero errors each time. 2 parity disks. 

 

Xeon X3440 with 16GB RAM.

 

 

 

 

Disks are either on a 3Gbs onboard SATA slot (or Esata but same port) or a 6GBPS backplane with H310 in IT mode.

 

So, the only thing I can figure is one disk is slow, but they never bench this slow. SLOWEST disk I have starts around or above 75Mbps but as you can see, it's super low above.

 

Ideas?

 

And I think I've done extended SMART checks on most/all the disks and SMART data isn't indicating much. A CRC error here and there but that's about it.

 

Some of the read/writes in the graph are a preclear. New-to-me drives that passed extended SMART and now I'm just doing a preclear before I 'bother' since they appear to be used but good shape datacenter drives. (Ebay. Warrantied, and I've got two brands / models for dual parity.)

 

The preclear is in a DS4246 on it's own controller. T310 Dell and the controller is in IIRC an 8X slot, so I don't think that's it.

 

Edited March 10, 2021 by RealActorRob

[JorgeB]

Possibly a disk having slower zone(s), use the diskspeed docker, it can sometimes find the problem if it's big enough. 

[RealActorRob]

7 hours ago, JorgeB said:

Possibly a disk having slower zone(s), use the diskspeed docker, it can sometimes find the problem if it's big enough. 

Yeah, I've used it but probably for best accuracy I need to shut down the array and test.

 

I've got 2x4TB undergoing preclear and will probably be replacing the slowest drives so hopefully that's one of them.

 

 

[PeteAron]

In my case i believe i have a failing drive but it's odd.  I am getting CRC errors even though i am sure it isnt the connections.  I mention it because i was using this drive to replace another disk in my array, and my rebuild speeds were varying from 10 Mb/s to the more normal 130-160 while trying to rebuild the disk.  I took it out of the array and replaced it with another.  

 

I am going to rma the drive, even though the only issue is crc errors - it is up to 1475 crc errors now.  You may have a similar issue.

 

Edit:  This drive had been precleared immediately before trying to put it into the array.

Edited March 10, 2021 by kimifelipe

[RealActorRob]

One thing I do see, it's not the 3TB drive I don't think because after 2TB it ramps up a LOT.

 

 

But it sucks at many points UP TO 2TB so I can't point the finger definitively at one of the other drives. 

 

 

So the last 1TB is only going to take 2 hrs. Almost 24hrs to check 2TB is ridiculous.

 

It doesn't seem like full drives would cause that much slowness during a read as nothing is being moved. Yes, I need to work on that with deletions/unBalance.

 

[itimpi]

21 minutes ago, RealActorRob said:

It doesn't seem like full drives would cause that much slowness during a read as nothing is being moved. Yes, I need to work on that with deletions/unBalance

how full a drive is will have no effect on a parity check as the parity check works at the raw sector level and is unaware of the meaning of the contents of the sectors - just that they have a bit pattern it is going to use/check.