[3836] in Athena Bugs
cron: network load spike every minute
daemon@ATHENA.MIT.EDU (probe@ATHENA.MIT.EDU)
Fri Dec 15 22:56:11 1989
From: probe@ATHENA.MIT.EDU
Date: Fri, 15 Dec 89 22:55:26 -0500
To: bugs@ATHENA.MIT.EDU, netreq@ATHENA.MIT.EDU
Cc: ops@ATHENA.MIT.EDU, athena-ws@ATHENA.MIT.EDU, rel-eng@ATHENA.MIT.EDU,
Reply-To: Richard Basch <probe@ATHENA.MIT.EDU>
After hearing from the Network Group that there is a load spike every
minute on the minute, I guessed that the cause was /etc/cron. Since
that prognosis, I have ran "netwatch" and discovered that every minute
on the minute, that there are several DIFFERENT hosts making NFS
requests to 18.72.0.15 (chiroptera). This adds support to the claim
that the cause is probably /etc/cron.
As a sidenote, 18.71.0.13 (hector) was also in the list of hosts making
quick connections. It would be interesting to see if hector had the
staff packs attached...
Recommendations for changes to /etc/cron include:
- Have it look for a crontab file locally. One possible method would
be to supply a command-line argument for the path in which to search.
Another would be to hard-code it. In this case, I don't care which
one is done.
- Have cron slew its requests by the part of the minute in which it was
first started. This will avoid other binaries from being paged in
at the same time, when an entry is found in crontab.
If there are no objections, I will do the following for the summer
release:
1) Have cron look in /site/usr/lib, instead of /usr/lib.
2) Slew the processing of crontab until the part of the minute at which
cron was first started. Differences in the boot sequence and fsck
times should cause a wider spread and lower loads on the network.
-Richard
