Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: RAID 5 vs RAID 10 benchmark
Jonathan Lewis wrote:
Note in-line
[Jonathan: can you please check out your posting style in one or two news clients. It plays merry hell with Thunderbird, for sure]
If you consider a single stripe across the traditional 5-disc raid,
then
either
(a) the dead disk is the parity disk (one time in 5) - in which case reads
don't do any extra reads, and you no longer have a write overhead on the other four. or (b) the dead disk is one of the data disks (4 chances in 5) - in which case three of the possibly reads won't have to do any extra reads, and only one read in 4 will have to read the other three disks and theparity
to infer re-create the fourth set of data.
"The real concern with RAID 5 is what happens when you pull one of the disks out. A single physical I/O *TO THE FAILED DISK* now requires a read to be made against every surviving member of the array, so that the missing data can be deduced."
And yes, that still misses out your case (a), but so did Cary in his article on RAID 5 at www.baarf.com, so I feel I'm in OK company with that one.
Do the extra four words mean I'd change my advice? RAID 5 is still lousy under failure conditions, that's the point. So no, they wouldn't.
I was careful to represent the issue as a 'concern', not a no-vote, however, and for SMEs, RAID5 might still represent the best of a difficult compromise.
Awaiting the wrath of BAARF for that one...
Regards
HJR
Received on Fri Nov 26 2004 - 14:06:59 CST