>>>>> "Edward" == Edward Ned Harvey <[EMAIL PROTECTED]> writes:
>> RAID can be implemented in hardware or software. Can't speak to all >> hardware >> RAID controllers, but NetApp does RAIDDP, and ZFS has RAIDZ2, both of >> which >> are double-parity-disk RAID (i.e. Survive 2 disk failures rather than >> 1). Edward> This is very interesting. Is there any hardware controller Edward> that does the same? I'm sure there are HW RAID controllers that do RAID6, but you need to be even more paranoid for true failure protection. In the extreme case of Netapps, you would have each shelf of 14 disks be on seperate controller, with no more than two shelves per-controller. Then, if you lose a controller, you don't lose any volumes, since the other controllers can just take up the slack. The idea is, of course, that you don't have any single point of failure. In fact, NetApps can have dual redundany loops, so you can have multiple paths to a disk via different controllers. So if a controller dies, the data is just re-routed. This is used most often in clusters of course, where each head has a controller for each shelf in the cluster. Edward> I do think there's value in the hardware raid controller. Edward> Guess what, we've all seen sometimes kernel crashes. And if Edward> we had enabled write caching, we'd encounter FS corruption for Edward> whatever had not yet been written. But with a caching raid Edward> card, with battery backup, you get all the performance benefit Edward> (if not more) with lower risk of data corruption. Edward> The performance difference is most apparent when working on a Edward> zillion small files. The most dramatic difference I've seen Edward> is when formatting the FS, because of course, that's such a Edward> sparse write operation. To format 1Tb on my Dell, ext3, it Edward> took 40 minutes without write cache enabled on the raid card, Edward> and it took 40 seconds with the write cache enabled. >From what I recall, Netapp doesn't use an HW raid controllers at all, just plain Fibre Channel controllers. The big thing they do hardware-wise is to have a honkin bug battery backed NVRAN card in the box, which holds all writes until they are committed to disk. If the system crashes or loses power, the battery keeps the data safe in NVRAM until reboot and then writes it out to make sure it's all consistent. I'm not sure if the StoreVault boxes have this feature. We use NetApps at work, but larger ones since we have big data needs. I just wish they would increase the size of aggregates back 16tb to make life simpler. John _______________________________________________ bblisa mailing list [email protected] http://www.bblisa.org/mailman/listinfo/bblisa
