Subject: Re: sucky performance on i386/1.3I
To: None <firstname.lastname@example.org>
From: R. C. Dowdeswell <email@example.com>
Date: 01/22/1999 16:59:58
On 917042286 seconds since the Beginning of the UNIX epoch
"Neil A. Carson" wrote:
>It seems like the scheduling is just going completely ga-ga.
This is a bit unrelated, but: I noticed that on NetBSD/alpha -current
as of too long ago that scheduling appeared to be broken.
(Unfortunately top was also broken, so I wasn't very precise at
noticing.) But when I was running a couple of CPU-only apps, here's
what I get after an extensive search for my 4.4BSD book (didn't
$ ps alux | grep -E rc5\|a.out
UID PID PPID CPU PRI NI VSZ RSS WCHAN STAT TT TIME COMMAND USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND
500 15849 15848 255 105 19 1128 488 - RN E0 5565:20.29 ./rc5des rc5 15849 47.0 0.8 1128 488 E0 RN Mon10AM 5565:20.29 ./rc5des
666 29272 29223 255 91 0 32 320 - R p1 9:38.49 ./a.out elric 29272 47.1 0.5 32 320 p1 R 3:07PM 9:38.49 ./a.out
Sorry for the long lines, just wanted to make sure that niceness
and CPU usage were included. As you can see rc5des is niced to 19,
and ./a.out isn't, but after 9:38.49, they both have about the same
CPU-usage. This suggests that something is broken.
(Just for refernce, a.out is:
I am going to sup/rebuild and will send-pr a bug if this problem
still exists on my machine.
I've been thinking about the scheduling a bit, since I am always
running rc5des, and I am a bit unsatisfied with the amount of CPU
a nice 20 proc uses (on the older versions of NetBSD that I've
used, where it works.)
== Roland Dowdeswell