[lkml]   [1998]   [Dec]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectSMP 2.1.131: SCSI performance extremely poor vs. IDE
I've received a report of extremely poor SCSI performance in the late
2.1.x kernels and verified this under 2.1.131.

I tested this by untarring the mozilla source onto the local hard
drive, syncing, and rm -rf ing it.

The mozilla source is 91.6 megabytes untarred.

The results are astounding:

[ben@gilgamesh:/tmp]% time rm -rf mozilla 4:15PM
rm -rf mozilla 0.13s user 0.93s system 18% cpu 5.581 total

[ben@beowulf:/tmp]% time rm -rf mozilla 4:15PM
rm -rf mozilla 0.17s user 1.51s system 98% cpu 1.712 total

This is, frankly, terrible. Since when is IDE three times faster than
ultra-wide SCSI?

The first computer, gilgamesh, is a SCSI-based Pentium II/333 system;
using the aic7xxx driver version 5.1.4/3.2.4 for its Adaptec AHA-294X
Ultra Wide SCSI controller to control an IBM DDRS-39130W ultra-wide

The second computer, beowulf, is an IDE-based dual Pentium Pro/200
system, PIIX3 IDE controller, ST39140A IDE drive.

Both are compiled with SMP = 1.

One would think that the SCSI Pentium II/333 system, even though it's
single-processor, would have decent disk performance! Is there
something I'm missing?

So as to not spam the list with the .configs of the two machines,
I've put them up on the Web for your perusal:

What would cause this kind of performance loss?

For an interesting comparison, here's the same test on an UltraSPARC:

bash$ time rm -rf mozilla/

real 1.1
user 0.2
sys 0.7


Brought to you by the letters Z and X and the number 0.
"You have my pills!" -- Grampa Simpson
Debian GNU/Linux -- where do you want to go tomorrow?
I'm on FurryMUCK as Che, and EFNet and YiffNet IRC as Che_Fox.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:46    [W:0.045 / U:6.660 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site