lkml.org 
[lkml]   [1998]   [Nov]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: Something wrong with scheduler in linux 2.1.127/2.1.128
Date
H.J. Lu enscribed thusly:
> Hi,

> I have a strange problem with scheduler in Linux 2.1.127/2.1.128 on a
> Pentium UP. At certain time, I had to wait for more than 10 minutes to
> start a new process, like ls in shell. The existing running processes
> seem ok. I can still do things as long as I don't try to start a new
> process. I don't know what triggers the problem. There is no kernel
> messages. The machine load is unusually high. It acts as if the
> scheduler gets in a tight loop and wastes CPU time.

I think you are seeing what I've been seeing the the latest few revs.
It takes a long time for this to get triggered so I've been very slowly doing
a process of elimination to isolation when and where and how. It sometimes
takes as much as a day before the things starts acting up for me.

I see the same symptoms as you do but have been gathering other data.
Seem that an active running process is ok but starting processes are a killer.
Top shows 0% idle but 95% to 105% (yes greater than 100%) System time. User
time is in low single digits. My first sign of trouble is a rapid rise in
load average. This thing is normally getting triggered with something making
a high file system demand (usually when I resync my mailbox in elm) and the
load average takes a jump anyways. What happens in this case, though, is that
the load average rises above the 1.0 - 1.5 I normally see and may peak out
in 30-60 seconds at around 15 (NO THERE IS NO DECIMAL POINT IN THAT). Only
choice I have then is to try and kill processes and try and reboot. Running
processes seem almost impossible to kill, even though their CPU time is
clocking up rapidly. With the high "system" percent, I would assume that
they are all looping around down in the kernel.

Something strange was going on with the schedular around 2.1.127pre3.
I could not compile sched.c if SMP was disabled but could if SMP was enabled.
With 2.1.127pre7 I could once again compile with SMP disabled, but this
problem suddenly cropped up... I thought it was something to do with the
experimental Samba code I was working with (I'm a member of the Samba Team).
I eliminated that and then thought it might be someone attacking my sendmail
(I typically end up with 3 or 4 sendmail processes eating time). Then I
thought it was insufficient inodes... Now I suspect something down in the
schedular like you...

Now for the history...

2.1.126 - Problem not seen - SMP disabled
2.1.127pre2 - Problem not seen - SMP disabled
2.1.127pre3 - Not seen but sched.c bombs gcc with segv with SMP disabled
2.1.127pre7 - Can compile with SMP disabled, problem first noticed...
2.1.127 - Problem present - SMP disabled, SMP enabled not tested.
2.1.128 - Problem present - SMP disabled, SMP enabled under test NOW.

So far... I have not seen this occur even one single time in a
build with SMP enabled. I've only been running 2.1.128 with SMP enabled
for a few hours at this point. I will post a follow-up, yeah or ney,
tommorrow. By this time tommorrow, if it bombs out again, it ain't the
SMP disabled stuff... If it doesn't bomb out again, I think something got
introduced around 2.1.127pre3 that didn't get properly fixed between that
and 2.1.127pre7 that has the SMP disabled case with its shorts in a knot...

I know the SMP disabled case is getting substantially less testing
than the SMP enabled case.

BTW... I'm testing this on two processors... One is a 200Mhz
Cyrix 686 and the other is a 200 MHx Pentium II. The problem has been
showing up on the Cyrix processor, but the Pentium has only been run on
2.1.127pre3 SMP and earlier. I just put 2.1.128 on the Pentium II but
haven't hit it with enough load yet to tell anything.

> --
> H.J. Lu (hjl@gnu.org)

Mike
--
Michael H. Warfield | (770) 985-6132 | mhw@WittsEnd.com
(The Mad Wizard) | (770) 925-8248 | http://www.wittsend.com/mhw/
NIC whois: MHW9 | An optimist believes we live in the best of all
PGP Key: 0xDF1DD471 | possible worlds. A pessimist is sure of it!
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:45    [from the cache]
©2003-2011 Jasper Spaans