raid5 autoselecting a slower checksum function

raid5 autoselecting a slower checksum function

[Meelis Roos]

This is 2.6.0-test2 on a Celeron 900:

raid5: measuring checksumming speed
   8regs     :  1640.000 MB/sec
   8regs_prefetch:  1316.000 MB/sec
   32regs    :   824.000 MB/sec
   32regs_prefetch:   788.000 MB/sec
   pIII_sse  :  1744.000 MB/sec
   pII_mmx   :  2244.000 MB/sec
   p5_mmx    :  2400.000 MB/sec
raid5: using function: pIII_sse (1744.000 MB/sec)

Why doesn't it select p5_mmx if it is 37% faster than pIII_sse?

-- 
Meelis Roos (mroos@linux.ee)

Re: raid5 autoselecting a slower checksum function

[Bruce Harada]

On Mon, 28 Jul 2003 18:17:00 +0300 (EEST)
Meelis Roos <mroos@linux.ee> wrote:

<snip>

> Why doesn't it select p5_mmx if it is 37% faster than pIII_sse?

This has come up before - see :

http://hypermail.idiosynkrasia.net/linux-kernel/archived/2003/week01/1894.html

Re: raid5 autoselecting a slower checksum function

[Tomas Szepe]

> [bharada@coral.ocn.ne.jp]
> 
> On Mon, 28 Jul 2003 18:17:00 +0300 (EEST)
> Meelis Roos <mroos@linux.ee> wrote:
> 
> <snip>
> 
> > Why doesn't it select p5_mmx if it is 37% faster than pIII_sse?
> 
> This has come up before - see :
> 
> http://hypermail.idiosynkrasia.net/linux-kernel/archived/2003/week01/1894.html

Fair enough, but wouldn't it be more appropriate if the kernel printed
a message like "SSE present, good.  No need to try the other checksumming
methods" in this case?

-- 
Tomas Szepe <szepe@pinerecords.com>

Re: raid5 autoselecting a slower checksum function

[Bruce Harada]

On Mon, 28 Jul 2003 22:54:38 +0200
Tomas Szepe <szepe@pinerecords.com> wrote:

> Fair enough, but wouldn't it be more appropriate if the kernel printed
> a message like "SSE present, good.  No need to try the other checksumming
> methods" in this case?

As I recall, someone actually posted a small patch to indicate that - I assume
it got dropped or lost. Certainly, the current message is rather confusing.