The problem with RAID-5 or RAID-6 is not the normal speed operation, it's the degraded performance when there is a drive failure. This includes read-only scenarios. A DB server getting any kind of real use will effectively appear to be down to client apps if it loses a drive from that RAID set.
Basically, think of RAID-5/6 as RAID-0 but with much slower writes, and a way to recover the data without going to backup tapes if there is a disc loss. It is NOT a solution for staying up in case of a failure.
Presumably, there is a business reason that you're thinking of using RAID-5/6 with hardware RAID and maybe a hot spare, rather than software RAID-0 which would save you 2-3 spindles of formatted capacity, plus the cost of the RAID card. Whatever that reason is, it's also a reason to use RAID-10.
If you absolutely need it to fit in 2U of rack space, you can get a 2U server with a bunch of 2.5" spindles and with 24x 500GB SATA you can get the same formatted size with RAID-10; or you can use an external SAS expander to put additional 3.5" drives in another enclosure.
If we're taking rackmount server RAID card votes, I've had good experiences with the LSI 8888 under Linux.
Cheers
Dave
On Tue, Nov 24, 2009 at 11:23 AM, Matthew Wakeling
<matthew@flymine.org> wrote:
We're about to purchase a new server to store some of our old databases, and I was wondering if someone could advise me on a RAID card. We want to make a 6-drive SATA RAID array out of 2TB drives, and it will be RAID 5 or 6 because there will be zero write traffic. The priority is stuffing as much storage into a small 2U rack as possible, with performance less important. We will be running Debian Linux.
People have mentioned Areca as making good RAID controllers. We're looking at the "Areca ARC-1220 PCI-Express x8 SATA II" as a possibility. Does anyone have an opinion on whether it is a turkey or a star?
Another possibility is a 3-ware card of some description.
Thanks in advance,
Matthew
--
Now you see why I said that the first seven minutes of this section will have
you looking for the nearest brick wall to beat your head against. This is
why I do it at the end of the lecture - so I can run.
-- Computer Science lecturer
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
