On 02/19/2014 09:18 AM, Edward Ned Harvey (lopser) wrote:
>> From: [email protected] [mailto:discuss-
>> [email protected]] On Behalf Of Skylar Thompson
>>
>> It really depends - while 24 10K disks will give you more concurrency
>> and throughput, it won't be able to beat the single-operation latency of
>> 15K disks.
> Basically, if you're going to benchmark a single operation, it doesn't matter
> what kind of disks you have, or how the raid is configured. A single disk =
> a mirror = raid5 =raid6 = raid10. (The only choice that will matter is SSD
> vs HDD.)
>
> But even if you have a bunch of random IO operations that are all independent
> of each other, they are NOT each an instance of a single operation. Drives
> and controllers have a lot of intelligence built into them, to optimize disk
> performance over a large pool of requested operations. If you give it 20
> random seeks concurrently, it will reorder those seeks using an elevator
> algorithm (or whatever algorithm it determines to be best) so the results
> will be returned not necessarily in the same order they were requested, but
> in the minimum possible time.
>
> I can't really think of a situation where you would care about the
> performance of a single operation. Meaning, if you *consistently* issue a
> single operation, and then wait for it to complete before you issue any more
> operations of any kind.
I agree with this. As an example, when some folks were benchmarking DB2
over DRBD, the I/O operations were 30% slower than local operations.
But the throughput in the worst-case (incredibly write-intensive)
benchmark was only 10% slower.
Database performance is much more than just "how fast are your disks".
--
Alan Robertson <[email protected]> - @OSSAlanR
"Openness is the foundation and preservative of friendship... Let me claim
from you at all times your undisguised opinions." - William Wilberforce
_______________________________________________
Discuss mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss
This list provided by the League of Professional System Administrators
http://lopsa.org/
