On Monday,  3 February 2003 at  9:45:00 +0200, Jukka Marin wrote:
> On Sat, Feb 01, 2003 at 11:08:32AM +0200, Antti Kantee wrote:
>> On Sat Feb 01 2003 at 10:04:25 +0200, Jukka Marin wrote:
>>> I have no experience of the RAID controllers, but I'm not particularly
>>> happy with the NetBSD RAIDframe implementation.  I bought fast disks,
>>> configured RAID - and found out that RAID drops the performance to a
>>> fraction of that of the disks.
>>>
>>> I don't know why, because when RAID is 100% busy, CPU load is almost
>>> zero and the disk load 25% or so.  It's like if RAIDframe had some
>>> usleep() calls in the code to make things go slower..
>>
>> What's your stripe unit size? I also was suffering from inexplainable
>> slowness until I dropped the stripe unit size to 16 sectors. Before that
>> my RAID5 gave something like 5MB/s write speeds, now it's giving more
>> than 25MB/s.
>
> Does this apply to RAID1 as well?

Yes.

> I understand that software RAID is slower than the disks (although
> with the CPU's of today...)

Not necessarily.

> but I do NOT understand why RAID is 100% busy when none of the
> resources of the server (CPU, disks, memory) is.  I see this on my
> server and a friend sees it on his.

The real issue is disk latency.  In fact, the disks probably *are*
100% busy, but up to 99% of that could be latency, especially if the
stripes are so small that you have to spread a transfer across all
disks.

Greg
--
Finger grog@lemis.com for PGP public key
See complete headers for address and phone numbers