httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From dean gaudet <dgaudet-list-new-ht...@arctic.org>
Subject Re: [PATCH] simple hash table implementation
Date Wed, 03 May 2000 16:18:49 GMT
On Wed, 3 May 2000 rbb@covalent.net wrote:

> 
> > >SIGBUS or SIGSEGV works fine for me.
> > 
> > At the moment the pools code calls a user-supplied abort function when
> > it fails to get any memory.
> 
> Actually, in Apache, the pools code just reports an error message and
> dies.  Although it is easy to repalce that with a different abort
> function.

either way, i'm a strong believer of putting enough RAM into your machine;
and benchmarking your service to figure out the proper MaxClients setting.
(heck you don't even need to do extravagent benchmarking, just turn your
service on, check the SZ/RSS of each process, the free ram, and do a
little math.)

'sides, linux with optimistic memory allocation won't ever return NULL
from malloc anyhow, you don't find out until you hit a SIGBUS that the
system doesn't have enough RAM.

and before the pessimistic memory allocation folks pipe up, i'd just like
to point out the wasted 2 to 4Gb of swap disk on all of my solaris servers
just because solaris insists on there being a backing page for every page
forked.  i'd never let a web server swap anyhow, so that disk is total
waste.

and so, i tend to think memory allocation failure checking is a waste of
cycles :)

-dean


Mime
View raw message