Re: Cheap FC arrays? (was Re: 10g RAC: max performance & min cost with miSCSI?)

JEDIDIAH wrote:
> What I hear from storage industry types is that SATA isn't
> quite yet ready for prime time yet. The drives aren't reliable enough
> and TCQ is either not there or not good enough.

The drives do have a worse MTBF than SCSI or FC drives. I was thinking of putting Oracle ASM into good use by striping and mirroring everything and running the RAID array JBOD mode. In a case of a disk malfunction ASM should be able to detect it automatically and rebalance the whole set of disks so that data is again redundant (2 copies of each data block). After the rebalance another disk could fail and the rebalance happens again on background as long as there is enough capacity to store all data twice on the remaining disks. And if the disks are 500 GB ones the data should fit easily. The question still remains how good Oracle ASM is a) in detecting disk failures and b) doing the rebalance fast enough to minimize the possibility for another disk failure to occur during the rebalance.

What comes to TCQ is that I would be putting high hopes on Apple RAID cache (2 x512 MB) & storage processors and on the fact that every disk on the array has a separate ATA controller. As stated in http://www.oracle.com/technology/deploy/availability/pdf/lcs_OW.doc.pdf a single array should be able to sustain 266 MB/s of read throughput on multi user sequential workload ("Data Warehouse").

--
Heikki