From owner-freebsd-hackers  Mon Feb  3 17:14:45 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id RAA04893
          for hackers-outgoing; Mon, 3 Feb 1997 17:14:45 -0800 (PST)
Received: from alpo.whistle.com (alpo.whistle.com [207.76.204.38])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id RAA04880
          for <hackers@freefall.freebsd.org>; Mon, 3 Feb 1997 17:14:38 -0800 (PST)
Received: from current1.whistle.com (current1.whistle.com [207.76.205.22])
          by alpo.whistle.com (8.8.5/8.8.4) with SMTP
	  id RAA22981; Mon, 3 Feb 1997 17:11:19 -0800 (PST)
Message-ID: <32F68C4F.3F54BC7E@whistle.com>
Date: Mon, 03 Feb 1997 17:09:35 -0800
From: Julian Elischer <julian@whistle.com>
Organization: Whistle Communications
X-Mailer: Mozilla 3.0Gold (X11; I; FreeBSD 2.2-CURRENT i386)
MIME-Version: 1.0
To: Bill Kish <kish@browncow.com>
CC: hackers@freefall.freebsd.org
Subject: Re: kernel malloc options
References: <199702031942.OAA18041@browncow.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

Bill Kish wrote:

>  I'm working on a a Network related module which performs a function
> similar to Network Address Translation. This module needs to
> dynamically allocate small (~40 byte) structures to keep track of
> currently established TCP connections between machines on the "inside"
> and "outside" networks.

Wow, a couple of months ago we had no address translation, now we
have *4* different ones.. :)

> 
>  Currently, I'm malloc()ing these structures as needed using the
> following parameters:
> 
>  malloc((u_long)sizeof(*conxn), M_TEMP, M_NOWAIT)
> 
>  This appears to work fine in my initial lightly loaded tests,
> however, I'm concerned about what will happen when the system is under
> heavy load. I have the following questions and would appreciate any
> advice offered!
What I do is cache a certain number of these..
I keep a free_thing list which
will hold a maximum of N items

If I ever free more than N then I start actually doing  "free(x,
M_TEMP)",
but until then I just keep them locally, this usually means that if the
number of
items is gradually increasing I'm doing malloc's, but with 
small scale fluctuations, they are satisfied from the cache.

of course malloc/free are not all that expensive
in general..


> 
>  1) Is malloc() a reasonable allocator for small highly dynamic
>    objects?
> 
>  2) Is M_TEMP a good choice? Does it make a difference?
it's fine. to use any other, you'd have to edit malloc.h
and for LKM stuff that's not an option.

> 
>  3) Are there any config options that affect the amount of pinned
>     kernel memory available for malloc? The system will be doing
>     little else besides dealing with these packets...
> 
> 
> 
>                 Thanks In Advance,
> 
>                   -BK
> --