Major issue with 9286cv-e?

D

danswartz

2[H]4U
Joined
Feb 25, 2011
Messages
3,715
So I got one of these a couple of days ago, with BBU. Threw 6 SAS 1TB drives on in RAID6. Nice speed (530MB/sec read/write). I decided to add another 1TB drive, and it started reconstruction. Glacially slow. What is far, far worse is that the controller is apparently so bogged down, that the iSCSI hosts I was serving storage to (two vsphere hosts), went APD and all my VMs crashed. It's so slow, in fact, that even typing a MegaCli64 command on the storage appliance takes like 30 seconds to even reply. I started the rebuild around 11AM and it's only now at 15%! It is set at the default 30% background rebuild rate, so this seems very wrong. Assuming it finally finishes the rebuild (giving me 1TB more), I could just be happy, but I am not willing to have this happen again, if I should need to do a rebuild. e.g. if a drive dies, I now have NO confidence I will be able to rebuild the array without the same crap happening. So, I would need to pull everything off the drive onto another pool first. If I'm going to do that anyway, I'd be better off just nuking the pool and rebuilding from scratch, no? This sounds incredibly lame! If I can't get a helpful answer from somewhere, this POS is going back to the seller :(
 
Reading intel notes they say that raid6 rebuild impairs data drive performance. Yeah, you can say that again! I am tempted to switch to raid10. I did try that initially, but the performance was about 20% worse than raid6 (remember this is with a BBU and write-back cache), so raid6 had 4 data drives, whereas raid10 only had 3? I am currently serving up a single large LUN to vsphere for VMs. Another alternative: create two 3-disk raid0 virtual disks, and mirror them in a zfs storage appliance. I prefer raid10 to raid 0+1, but... Yet another possibility (which I don't know is feasible yet or not): Put the controller in JBOD mode. The concern there is that those drives might not be able to take advantage of the write-back cache. If not, there is little reason to keep this controller, and I may as well return it for a refund, and put my old 9207 HBA in and go back to vanilla zfs :( I guess one more experiment I can do, once I get everything off the array: Rebuild it as raid10, then initiate a rebuild and see if it goes faster. If it does, might be worth going that route...
 
Dan- What is the underlying OS? How much pounding is the box getting to the VM's while doing the rebuild? It sounds like something is wrong with that card unless it is a settings issue. Though I generally recommend Areca cards, the 9286 is a current card with a relatively fast SOC.
 
OS is linux 3.14 (ESOS storage appliance). Very little pounding. At the moment, it is servicing a single gigabit iSCSI link. It was sailing along just fine until I added the 7th drive to the array, then the wheels came off. It's still resynchronizing just fine (it will be done sometime tomorrow), but it seems to be virtually catatonic when it comes to talking to the outside world. I'm giving serious thought to contacting scsi4me (where I bought it) and ask for an RMA.
 
mwroobel, do you have any specific recommendations? i had thought raid6, since the read/write perf was very good (due to the writeback cache?), but from what i have read, the rebuild time will be much longer than for raid10 (where it only needs to rebuild one drive). i was thinking areca 1680/1880/1882?
 
You must log in or register to reply here.
Back
Top