I'm working with a machine that's operating as a NAT router and recursive DNS
resolver and is also running the Squid disk cache. Squid, in turn, spawns the
"diskd" daemon, which does disk accesses on behalf of Squid. When Squid
spawns diskd, it gives it a priority level 6 greater than itself. In other
words, if Squid is launched normally, it gets a priority of 2 (normal) while
diskd gets a priority of -4 (very high).
Unfortunately, diskd is not an efficient user of CPU (it seems to be polling
for I/O completion) and is starving other processes on the machine (for
example, natd) which need to operate in near real time.
I'd like to keep diskd running on that machine, because having disk access
done by a separate process is very efficient -- even more so if the system
uses SMP. But I need to re-prioritize Squid and diskd to keep the rest of the
machine functional. In particular, I'd like to nice Squid down by 1 (so that
natd and named have priority over it) and have diskd run at standard priority
(so that it can't starve other processes). This will keep diskd at a higher
priority than Squid itself, which in turn will hopefully prevent message
queues from overflowing.
Reducing Squid's priority is simple; I can just edit the script that starts
Squid so that /usr/bin/nice is used to invoke it. But taming diskd is more
difficult, because diskd is a child process of Squid. I have to make sure it
has started (which may require a delay loop), find out its PID, and then
"renice" it by whatever increment is required to get it to the system's
standard priority (2 by convention). Is there a "renice by name" utility for
FreeBSD (sort of an equivalent of "killall")? I could gin one up, but since
this seems like something that people would want to do frequently, find it
hard to believe that someone hasn't already written one.
Google is your friend :)
very first link.
email@example.com mailing list
To unsubscribe, send any mail to "[EMAIL PROTECTED]"