SUMMARY: Access to systems with high load.

From: Charlie Dennett (dennett@Kodak.COM)
Date: Wed Oct 23 1991 - 11:06:07 CDT

My original query:

----- Begin Included Message -----

At this moment I have a 4/490 server with a load you wouldn't believe.
Here it is (it was 20 points lower about a half hour ago):

    sunshine up 13 days, 2:44, load average: 261.47, 258.00, 250.99

I am not currently at the same physical location as the server. Is there
anyway possible of accessing a system in such a state? Is there some
hidden flag for telnet or rlogin that allows root to login with a
-19 nice value? (Other non-UNIX systems I worked on in the past had
such an access method.) If there is a way it probably won't help this
time but just may save my butt next time.
----- End Included Message -----

Some additional background. The above load figures were around 1830
hours. By 0430 the next morning the load was in the 880 range (1
minute load). The system stopped responding to my job running on
another machine that was doing a `rup sunshine' once a minute and
storing the output in a file. Yes, it answered to rup and all nfs
traffic to and from sunshine seemed normal. Rlogin and telnet would
not work, however.

I don't know what started the load climbing but I know what added to
it. I have a program I obtained from Usenet that monitors user
processes and exec's off renice commands for processes that use too
much cpu time. Apparently the renice commands where stalled and caused
the load to build up. Evidence of this was from a logfile the autonice
program maintains. It kept trying to renice the same user process once
a minute to the same nice value. This indicated that any previous
attempt to renice the process had not yet taken effect when the program
woke up a minute later.

Anyway, I recieved two suggestion and several "let me know what you
find out" messages.

The first suggestion came from a collegue when I called him at home.
He suggested adding the string `pri=-20*' to the beginning of my GCOS
field in my password entry. That didn't work. My collegue is trying
to research this and discover where he found out about it.

The other suggestion was to try:

        rsh -n sunshine nice -19 ps -auxwww

This didn't work either.

Oh well, thanks to those who replied. I appreciate the effort. Perhaps
I can return the favor some day.

Charlie Dennett | Rochester Distributed Computer Services
Mail Stop 01816 | Customer Technical Site Services
Eastman Kodak Company |
Rochester, NY 14650-1816 | Internet: dennett@Kodak.COM

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:17 CDT