On Mon, Feb 26, 2007 at 08:09:57PM -0800, Steven Kurylo wrote:
> On 2/26/07, Les Klassen Hamm <les@bitlink.ca> wrote:
> >Greetings all,
> >
> >Can someone point me to some good info on hard drive reading speeds?
> >More specifically, if software needs to do a great deal of reading from
> >lots of different files, how would a pair of 7200rpm drives in RAID 0
> >compare to one 10K rpm drive?
>
> I'm going to assuming you meant raid 1. As well relatively small random
> seeks.
>
> I would say any given single transaction will be faster on the single
> drive. However multiple transactions will be faster on the raid 1
> because the linux kernel can balance the requests across the drives.
>
> Of course it would be interesting to test.
For sure. I think milage varies depending on the specific implementation
of both the RAID and the hardware set-up.
For example, a software (or pesudo-hardware RAID) from 2 P-ATA drives in a
master-slave arrangement wouldn't take advantage of parallelized reads since
the P-ATA bus would block, depending on the driver implementation could also
block but this is more common with ATA/33 and slower.
In general, for many small read transactions, more spindles would probably
be better. This assumes that the RAID controller or software is able to
parallelize operations effectively. I recall reading back with my first
Promise RAID controller years ago that RAID 0 would improve performance over
a single drive, I'm sure that info is somewhere in the Promise documentation
archives. Basically being able to parallelize the seeks would speed the
system up.
If you're considering the trade-off, a risk/exposure assessment of the two
options is important. With a RAID 0, the risk of a drive failing is higher
since P(array failure) = 2 x P(single drive failure). You lose data if
either drive (in a 2 disc RAID 0 array) fails. I think that's why Steven
is questioning the RAID 0 choice.
Personally, I'd go with the single disc if I cared about reliability or go
RAID >0.
Received on Tue Feb 27 01:31:40 2007
This archive was generated by hypermail 2.1.8 : Tue Feb 27 2007 - 01:31:47 CST