[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: good value for MAXUSERS on big server?
Quoting David Wilk (firstname.lastname@example.org):
> I've been trying to figure out what significance MAXUSERS has on a larger
> system. It seems that it may affect performance with regard to large
> filesystems and large numbers of processes.
> I'm curious if anyone can shed some light on this kernel parameter?
> thanks for the help.
Hal Stern (Sun, author) once referred to MAXUSERS as a "general
volume knob" for sizing various tables.
You use the term "larger system" from which I can infer that:
- it takes up a lot of room
- its used by hundreds or thousands of users
- it has lots and lots of RAM
- there are many many disks on it and/or bazillions files on it.
Or combinations. Not really specific.
OReilly has a dated, yet pertinent book I got in 1993 or so called
System Performance Tuning. I still refer to it.
It covers 4.2/3 BSD and System V but frankly, the principles
still are relavent. And many of the variables and tools have
the same names. Many of them are brought out in the sysctl(8)
tree these days.
I had learned the rule of thumb of setting MAXUSERS to the number
of MB of RAM you had. Mainly worked on a typical system. 96 on a
96MB system. Of course running a render farm or a database server
often required tweaking dials a little bit - more shared memory, a
little tighter scheduling, etc. On my 2GB systems, 2000 MAXUSERS
is nuts, the tables just don't need to be that large.
Learning to read the system during normal and also stressed use and
identifying what resources are limited is the job of a good system
admin (after adding users, filling toner, speaking management-ese
and replacing keyboards that users have spilled coffee into).
Reading for building better systems and stronger resumes:
System Performance Tuning - OReilly. Specifics dated; principles timeless
Sun Performance & Tuning - Adrian Cockroft of Sun. A little Sun
oriented, not really an issue. Good notions.
Design and Implementation of 4.4BSD - McKusick, Bostic, et al.
Details of 4.4 kernel circa 1996.
The Source Code - read previous and learn the major blocks of the kernel