Re: udp_outgoing_peer (was: udp_incoming/outgoing_address)

From: <jw@dont-contact.us>
Date: Mon, 05 Apr 1999 21:35:08 +0200

Hi Henrik,

> Now I see your problem. You do not have a sound TCP/IP network. You
> can't make a working configuration because you have the same IP address
> active on both nodes at the same time. What you really need is multiple
> ICP ports to allow Squid to listen for ICP queries on multiple
> addresses, or a smarter clustering method which does not have the same
> IP address active on both nodes at the same time.
>
Well, because it's not really a cluster but a failsafe implementation
using a dynamic routing protocol the latter is not possible.

> Squid currently supports peering with clients on one address and peering
> with parents on another. Extending this to allow receiving queries on
> multiple addresses is planned.
>
Well, at least this should be enough, because for the peering we may
use the hard interface address. And the clients may use the virtual
ones.
The only problem IMHO is, that any reply will get the hard interface
address from icpCreateMessage instead of using udp_outgoing_address.

> Another way to solve your problem would be to use different networks for
> the official (client announced) addresses and the machine addresses, and
> have client traffic routed throught the right announced interface. That
> way you can get the desired effect of multiple TCP ports without
> actually having Squid support it.
>
Well, because of the two different locations this would be to much
overhead just for sibling each other proxy. (we currently do have
a sibling hit rate of about 1 percent at 35 GB cachdir)

Well, we'll use my patch for now and the future feature mentioned above,
if available.

Thanks for the hints & time,

        Jan
Received on Mon Apr 05 1999 - 13:25:55 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:45:43 MST