8.04 super slow on dell 2550

[Jason Joines]

     I have a dell 2550 server that I want to get Ubuntu on.  I tried a 
few months ago with the initial 8.04 release and the install was so slow 
I gave up.  I had tried the mini.iso cd, desktop cd, server cd, and 
alternate cd with the same results.

     Recently I decided to try again with the 8.04.2 media and just let 
it keep running.  Instead of assuming it had timed out I would go check 
on it once a day and after many days it was finally installed.  As an 
example of the slow install, when using the text install and at the 
point where the dialog said "Installing the base system" it took 66 
minutes to go from a 33% status to a 39% status.  Also, by watching logs 
it seemed that the package retrieval from the internet was really fast 
but it became super slow any time it was "unpacking".

     The system has a broadcom1 Gbps nic, two pentium III 933 cpu's, 1 
GB of RAM, three 74 GB u160 scsi drives configured as raid 0 via the 
builtin dell raid controller which is using the aacraid module.  I've 
tried both ext3 and xfs as the filesystem.

     This system was running windows server 2k3 and wasn't noticeably 
slow.  I also have an identical system that is running Ubuntu 7.10 and 
it seems to be over three times as fast via these simple tests:

     7.10 System running 2.6.22 kernel
dd if=/dev/zero of=gigfile00 bs=1k count=1048576
1048576+0 records in
1048576+0 records out
1073741824 bytes (1.1 GB) copied, 29.5587 seconds, 36.3 MB/s

dd if=/dev/zero of=gigfile00 bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 29.1349 seconds, 36.9 MB/s

     8.04.2 System running 2.6.24 kernel
dd if=/dev/zero of=gigfile00 bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 98.4516 s, 10.9 MB/s

dd if=/dev/zero of=gigfile00 bs=1k count=1048576
1048576+0 records in
1048576+0 records out
1073741824 bytes (1.1 GB) copied, 93.9246 s, 11.4 MB/s


     I found this bug report 
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/249964 about these 
smp systems having the aacraid driver hang under high load.  However, 
this box isn't under any load and I haven't had any of the log messages 
reported.  I decided to try the fix anyway so I recompiled the kernel 
after changing:
#define AAC_MAX_32BIT_SGBCOUNT  ((unsigned short)256)
to
#define AAC_MAX_32BIT_SGBCOUNT  ((unsigned short)127)
in aacraid.  It didn't help.

     Any suggestions?


Jason Joines
=================================