We have a legacy MRP application that runs on Informix SE. Users log
in with SSH or telnet, and there may be as many as 100 terminal
sessions active at any time. We have been running this for many years
on Solaris with 2 x 400 MHz processors and 2 GBytes of RAM. We would
like to migrate to a modern SLES10 box with multiple 3 GHz processors
and 4+ GBytes RAM.

When users run reports on this application, they get pretty CPU
intensive. If multiple users simultaneously run reports, the
interactive performance gets really slow. Disk and memory usage are
fine. It seems that users in an IO wait may take up to 60 seconds to
get back into a run state. We have never seen this problem with the
same application on the old Solaris box. I am not a Linux internals
guy, and my theoretical knowledge would be from UNIX many many years
ago. My gut tells me that the process scheduler is favoring the
running processes over TTY sessions, which probably makes sense for a
normal server that is running mostly client/server apps. I can't do
anything about how the legacy applicatioon operates, but I would be
willing to favor interactive performance over calculation run time.

Can anyone point me to ideas on how to optimize this box for many TTY
interactive sessions that will perform well even when CPU intensive
processes are running? Thanks for any advice!