[488] in linux-net channel archive

home help back first fref pref prev next nref lref last post

Aliases and more

daemon@ATHENA.MIT.EDU (Alan Cox)
Wed Jun 14 07:27:56 1995

From: iialan@iifeak.swan.ac.uk (Alan Cox)
To: linux-kernel@vger.rutgers.edu, linux-net@vger.rutgers.edu
Date: Wed, 14 Jun 1995 09:56:58 +0100 (BST)

Traceroute is provisionally fixed, and a bug that stops 1.1.x and 1.2.x
hosts with the last byte of their address being over 223 has also been sent
to Linus as a fix.

I've submitted the changes to allow higher than default listen() calls (eg
for big apache based http sites) for both kernels.

I'm just tidying up the loadable firewall modules from Mazama to go in today
and then after that start hacking the new buffers that were put into a 1.1.4x/5x
kernel ages ago (when it was about to be 1.2/1.3 and Linus then went to 1.1.9x)
over the next week or so.

After that the alias stuff will get a look in.

>Another reason for a rewrite is that I don't believe Alan Cox will
>ever be happy with the linked list approach I'm using now, although I
>think it's no big deal for the common application.  However, it's
>*his* code I'm messing with, so I'll do it the way he wants it :-)

The reason for this is simple. A pentium fetches memory in 32 byte bursts
so chasing a linked list rather than array of pointers is woefully inefficient
and tends to generate excess reads and cache dirtying. Since you generally
add aliases once per boot and look them up n thousands times (or n million
for some sites) is better to do the slow kmalloc/copy/kfree cycle when adding
an alias.

I've got a fair number of patches queued up, but I've undoubtedly lost some
so do fee free to check in a few kernels time if something seems to have gone
missing.
 
Alan


home help back first fref pref prev next nref lref last post