Unraid woes, controller disappering

[porina]

I'll try to keep this short with relevant details.

 

Original build:

Z170 ATX mobo, a bunch of storage disks, one parity. All was fine apart from the case had bad airflow and I didn't like the temps.

Bought Node 804. Z170 was ATX so wouldn't fit. Swapped it with a mATX H81 mobo (Haswell). Added an extra controller card for more ports. Added two cache drives. System always missed some HDs on 1st boot, but they came back on a restart.

Swap the board with a mATX Z97. I just tried writing to it, but the controller card shows an IO error after a little use and I lose the cache drives. It was fine with the older mobo.

 

The controller in question is 6 SATA port, based on ASM1062+1093. That's PCIe 2.0. Since I'm only connecting HDs off it, I'm not concerned about the bandwidth. It worked fine for hundreds of GB on the H81 mobo but now the Z97 is giving me a problem. The only thing I can think of, which I'm about to try, is disabling spread spectrum, but I don't think there are many things to tinker with here.

 

Suppose worst case, I'll try a different controller as all I want are a bunch of dumb ports.

 

As far as I know the system is stable. I've even turned off XMP since it doesn't add value in this use case.

[AbydosOne]

30 minutes ago, porina said:

The controller in question is 6 SATA port, based on ASM1062+1093

I had trouble with ASMedia controllers in FreeNAS, so I wouldn't be surprised if it's just bad *nix support. Getting a IT mode LBA would be your best bet (something like an LSI 9207-8i or such, just remember to get breakout cables too).

[porina]

3 minutes ago, AbydosOne said:

I had trouble with ASMedia controllers in FreeNAS, so I wouldn't be surprised if it's just bad *nix support. Getting a IT mode LBA would be your best bet (something like an LSI 9207-8i or such, just remember to get breakout cables too).

I'll keep that in mind as an option.

 

I've disabled spread spectrum and while I was in bios, also set MCE to disable. Both options were "Auto" which I hate, since I have no idea if they are on or off. Anyway, since applying those changes I have written 90GB without the previous problem reappearing.

 

On the down side, now I have 4 UDMA CRC errors on one of the drives hanging off that controller. Not as bad as before, but still not good. Again, this drive, with the same cables going to the same card, was fine with the previous mobo. Maybe I'll change the cable and see what happens. I'm half tempted to just ditch the cache disks for now, which are the only drives on that controller.

[AbydosOne]

4 minutes ago, porina said:

Maybe I'll change the cable and see what happens

I've also had a bad cable cause CRC errors too, so yeah, try that first.

 

5 minutes ago, porina said:

I'm half tempted to just ditch the cache disks for now

I've done a lot of on-again-off-again research regarding L2ARC in Free/TrueNAS (a form of caching), and I could never really justify it. Unless you're running huge database loads, the average user either doesn't access enough of the same data often enough to warrant more than RAM caching, or the array itself could already saturate any connections to it. Not sure how much different it is in Unraid, but I suspect it's pretty similar.

[porina]

1 minute ago, AbydosOne said:

I've done a lot of on-again-off-again research regarding L2ARC in Free/TrueNAS (a form of caching), and I could never really justify it. Unless you're running huge database loads, the average user either doesn't access enough of the same data often enough to warrant more than RAM caching, or the array itself could already saturate any connections to it. Not sure how much different it is in Unraid, but I suspect it's pretty similar.

Very different. Unraid uses fixed parity disks, and to reduce load on inactive disks it will read-modify-write that parity data. This significantly reduces its write performance. Here, cache disks are used to get around that. They are primarily a write cache, since read performance is not affected by Unraid's system. I have two cache disks as a mirror for redundancy. Typical writes without cache may be only around 50MB/s, and with cache disks I can saturate gigabit ethernet easily. Using cache disks means I have to wait less for writes to happen, so they're not essential.

[porina]

Ok, switched a pair of new SATA cables over for the two drives connected to the controller card. While only one was reporting errors, might as well do both to save me figuring out which one of the pair it was. Anyway, since then, no further errors have been indicated. I might finally be at the end of what was supposed to be a simple mobo swap. If a mobo swap ever was simple... hopefully now I can pretty much forget about touching it again until such point I get a disk that needs replacing, or I feel a need to add more.