IDE performances, 2.4 vs 2.5

IDE performances, 2.4 vs 2.5

[Stefano Rivoir]

Noting that 2.5 is much slower than 2.4 on disk operations (you *touch* 
it when you have not-so-fast machine and use KDE, for example), I've 
written a silly test that fwrite/fread a single 100Mb file, char by 
char, and timing it I have results that I can't understand very well. Of 
course, same machine, same hdparm settings, same processes running 
(none, it's a notebook without server processes). I've run these test 
several time, the results are always more or less the same (ext2):

2.4.19

   read:    real    0m15.822s
            user    0m15.180s
            sys     0m0.270s

   write:   real    0m12.524s
            user    0m11.800s
            sys     0m0.690s

2.5.70 (up to -bk14, and -mm6)

   read:    real    0m20.790s
            user    0m14.372s
            sys     0m0.949s

   write:   real    0m13.148s
            user    0m11.901s
            sys     0m0.665

Writing does not drop, but reading has a 6 seconds difference between 
user+sys and real that I can't figure out. And the total difference is 
"huge". Actually, using anything that touches the disk (it can be a 
trivial "aptitude" loading the cache, or a complex KDE) slows down.

I've run these tests on a HP Omnibook w/Celeron, but I have the same 
slow down on a Athlon K7.

Is it anyway "normal", something I should expect upgrading from 2.4 to 
2.5/2.6? Or there should be something I should check more accurately?

Bye all.

-- 
Stefano RIVOIR

Re: IDE performances, 2.4 vs 2.5

[Bartlomiej Zolnierkiewicz]

IDE layer is basically the same in 2.4.21-rc and 2.5.70 (but not in -bk).
If you check 2.4.21-rc against 2.5.70 and results will be similar,
it means its not IDE performance problem, but block layer, VM or FS.

Regards,
--
Bartlomiej

On Tue, 10 Jun 2003, Stefano Rivoir wrote:

> Noting that 2.5 is much slower than 2.4 on disk operations (you *touch*
> it when you have not-so-fast machine and use KDE, for example), I've
> written a silly test that fwrite/fread a single 100Mb file, char by
> char, and timing it I have results that I can't understand very well. Of
> course, same machine, same hdparm settings, same processes running
> (none, it's a notebook without server processes). I've run these test
> several time, the results are always more or less the same (ext2):

fwrite/fread is not a good test for IDE performance.

> 2.4.19
>
>    read:    real    0m15.822s
>             user    0m15.180s
>             sys     0m0.270s
>
>    write:   real    0m12.524s
>             user    0m11.800s
>             sys     0m0.690s
>
> 2.5.70 (up to -bk14, and -mm6)
>
>    read:    real    0m20.790s
>             user    0m14.372s
>             sys     0m0.949s
>
>    write:   real    0m13.148s
>             user    0m11.901s
>             sys     0m0.665
>
> Writing does not drop, but reading has a 6 seconds difference between
> user+sys and real that I can't figure out. And the total difference is
> "huge". Actually, using anything that touches the disk (it can be a
> trivial "aptitude" loading the cache, or a complex KDE) slows down.
>
> I've run these tests on a HP Omnibook w/Celeron, but I have the same
> slow down on a Athlon K7.
>
> Is it anyway "normal", something I should expect upgrading from 2.4 to
> 2.5/2.6? Or there should be something I should check more accurately?
>
> Bye all.
>
> --
> Stefano RIVOIR
>
>
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: IDE performances, 2.4 vs 2.5

[john]

> Noting that 2.5 is much slower than 2.4 on disk operations (you *touch*
> it when you have not-so-fast machine and use KDE, for example)

I noticed that KDE 3.1.1 was noticably slower to start on 2.5.69 than 2.4.21-rc1
when I briefly tested it on an Athlon XP 2200+ with 512 MB of RAM.

(Unfortunaly I haven't had much time for 2.5 testing, or any testing actually,
for a few months, but I tested it briefly when I first built this machine,
before it went in to production use.  Once all the security issues are addressed
in the 2.5 tree, I intend to start using it on a few production boxes, though).

John.