On Fri, 2006-03-24 at 15:38 +1100, Neil Brown wrote:
> On Thursday March 23, [EMAIL PROTECTED] wrote:
> > Neil - Thank you very much for the response.
> >
> > In my tests with identically configured raid0 and raid5 arrays, raid5
> > initially had much lower throughput during reads. I had assume
Hello,
I profiled some raid5 reads using oprofile to try to track down the
suspiciously high cpu load I see. This uses the same 8-disk SATA setup
as I had described earlier. One of runs is on a 1MB chunk raid5, the
other on a 32MB chunk raid5. As Neil suggested memcpy is a big part of
the cpu l
On Sat, 25 Mar 2006, Alex Izvorski wrote:
> http://linuxraid.pastebin.com/621363 - oprofile annotated assembly
it looks to me like a lot of time is spent in __find_stripe() ... i wonder
if the hash is working properly.
in raid5.c you could try changing
#define stripe_hash(conf, sect) (&((conf
On Sat, 2006-03-25 at 21:38 -0800, dean gaudet wrote:
> On Sat, 25 Mar 2006, Alex Izvorski wrote:
>
> > http://linuxraid.pastebin.com/621363 - oprofile annotated assembly
>
> it looks to me like a lot of time is spent in __find_stripe() ... i wonder
> if the hash is working properly.
>
> in ra
On Sat, 1 Apr 2006, Alex Izvorski wrote:
> Dean - I think I see what you mean, you're looking at this line in the
> assembly?
>
> 65830 16.8830 : c1f: cmp%rcx,0x28(%rax)
yup that's the one... that's probably a fair number of cache (or tlb)
misses going on right there.
> I looked
On Sat, 2006-04-01 at 14:28 -0800, dean gaudet wrote:
> i'm guessing there's a good reason for STRIPE_SIZE being 4KiB -- 'cause
> otherwise it'd be cool to run with STRIPE_SIZE the same as your raid
> chunksize... which would decrease the number of entries -- much more
> desirable than increasin
On Sat, 1 Apr 2006, Alex Izvorski wrote:
> On Sat, 2006-04-01 at 14:28 -0800, dean gaudet wrote:
> > i'm guessing there's a good reason for STRIPE_SIZE being 4KiB -- 'cause
> > otherwise it'd be cool to run with STRIPE_SIZE the same as your raid
> > chunksize... which would decrease the number
