Re: File Descriptors & Linux 2.0

John Henders (jhenders@stdismas.bogon.com)
25 Jun 1996 14:10:18 -0700


In <199606251746.KAA08888@neteng.engr.sgi.com> "David S. Miller" <dm@neteng.engr.sgi.com> writes:

> From: matthew@crocker.com (Matthew S. Crocker)

> I have changed NR_OPEN & NR_FILES to 1024 in /usr/include/linux/??.h

>You can also echo numeric values into:

>/proc/sys/kernel/file-max
>/proc/sys/kernel/file-nr

And, you may have to ulimit the program you're running when starting it,
unless there's a more elegant way to do this under linux. One our BSDI
systems, we start named, INN, and httpd with shell scripts that do a
ulimit -n 255 -d unlimited then exec the program. The reason we limit
open files to 255 is that this is also used to limit the number of
sockets that can be opened, and BSDI, and Linux as well from what I've
heard, are limited by bitmap size to the number of sockets you can open
and you will run into problems if you use a higher number.

Apache is a great server but, the way it opens log files for each
virtual server limits the number of servers you can run considerably.
A possible solution would be to use the new configurable log format
module with apache 1.1b and write all your logs to the same file, then
split them out later.

INN dealt with this problem for sites feeding a lot of uucp sites by
developing the buffchan program, which runs as a seperate program and
explodes the input into a number of output files. Something similar
would be handy for a web server. I played around a bit with buffchan
listening on a pipe and having the web server writing to the named pipe
but didn't take it very far beyond proof of concept.

As to named, I can't see any way to limit the addresses named will
listen on. The only solution I've found is not to run your primary
nameserver on a web server machine. Because of the problems named has
with this and swapping, it seems you're far better off to give named
it's own machine, or at worst a very lightly loader one.

-- 
      Artificial Intelligence stands no chance against Natural Stupidity.
                GAT d- -p+(--) c++++ l++ u++ t- m--- W--- !v
                     b+++ e* s-/+ n-(?) h++ f+g+ w+++ y*