Transparent cache

From: Yar Tikhiy <yar@dont-contact.us>
Date: Wed, 22 Oct 1997 12:46:29 +0400 (MSD)

Hello everybody,

I've installed squid 1.1.14 on a FreeBSD box with NAT (hacked natd)
as a transparent cache for a relatively small (about a few thosands)
clients. So, I've got two questions.

1. Wouldn't it be better to use "Host:" header for URL completion
than the virtual proxy mode? Only outdated versions of Lynx do not
send the header now, so they cannot fetch virtual-hosted URLs and
must be upgraded even when there is no transparent cache.
And if you use the "Host:" header you can set up peering and your
cache will contain only one copy of each object instead of a few
if object's source web-server has a number of IP addresses.
At least, squid may have used a slightly different algorythm when
deciding how to form an URL.
It's so now (see icp.c):
        if (vhost_mode)
        {
                URL is formed using quad-dot IP address;
        } else if (opt_accel_uses_host && (there is "Host:" header))
        {
                URL is formed using "Host:" header;
        } else ...

IMHO, it would be better:
        if (opt_accel_uses_host && (there is "Host:" header))
        {
                URL is formed using "Host:" header;
        } else if (vhost_mode)
        {
                URL is formed using quad-dot IP address;
        } else ...

What could you say about that?

2. As I could not set up the cache on our router (it is cisco :-)
I've run into trouble with ICMP messages of type 3 code 4 (Need Frag)
as they are neccessary for TCP protocol operation, but had aliased
addresses and went to the "real" destination host (e.g. www.microsoft.com :-)
instead of the cache server. Have anyone overcome the problem in a nice way?
(If it's interesting to anybody I can tell about my solution)

SY, Yar
Received on Wed Oct 22 1997 - 01:48:20 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:19 MST