Re: kernel tuning and MAXUSERS

[chuck <chuck@Yerkes.com>]

Quoting Matthew Patton (matthew.patton@ra.pae.osd.mil):
> I have a box that hosts, ftp/mail/pop3/imap4/http and soon innd or
> equivalent. I wanted to look into tuning it's performance a little mainly
> by tweaking some kernel constants. I defined MAXUSERS to be higher than 64
> and config complained. Is there not much point in going higher than that
> value?

Hal Stern once offered "Think of MAXUSERS as a overall volume knob."

Generally (*generally*) as a guideline, I've used 1 MAXUSER/mb - so
a 64 meg machine gets Maxusers: 64.  This works okay for me on a
desktop workstation.

Its value is used to define sizes of various kernel tables.  You may
want to hand tune some of those table sizes.

For servers, you will likely want to monitor behavior and table
sizes and tune those directly.  It depends on what you serve and
what sort of volumes you get.  News servers (many files, lots of IO)
have different characteristics than web servers (many short-lived
TCP based hits).

A good reference I used was "System Performance Tuning" from
O'Reilly & Associates.  This was useful to me in 1994 or so, I don't
know how well it translates to 1998 and 44bsd, but the principles
are excellent.

chuck