IDE performances, 2.4 vs 2.5

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

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

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