I think most people know about the memory hogging programs. At least in
1.2.13 and much of the 1.3.x series, it is possible for the kernel to
kill system daemons if a user program does something funny. I've had
inetd and bdflush die on me due to a faulty program I wrote (run without
privledge).
Also I'm not sure if there has been stuff written or modified, but there
seems something can still be done with the scheduler/swapper. Again in
the afformentioned kernels, a cycle/IO hungry process can really impact
system performance. I was quite horrified comparing my Sparc clone
(SunOS 403) which had 8MB (12ms SCSI2 in SCSI1 mode HD) versus a 8MB
486/33 (10ms ISA IDE HD) ... Granted the 486 would be slower, but the
Sparc has a BIG problem with RAM since the kernel eats up 3 MB (whereas
Linux kernels take up around 1MB) However, SunOS works with swap much
more smoothly than the 486 box under heavy load (you can actually do
stuff on the sparc when the 486 is crawling).
Any ideas? Maybe this has been fixed, I have a better box now (a P5
dominates over my sparc so it wouldn't be a fair comparison), or maybe
it's just a limitation of the PC hardware subsystem?
Thanks for any ideas,
-bc