Patch for Squid-2.4DEVEL2

Hello,

Could you make patch available to Squid-2.4DEVEL2?

Thanks
Balgaa
---------- Forwarded message ----------
Date: Mon, 17 Apr 2000 21:42:58 +1000
From: Robert Cohen <robert@apex.net.au>
To: squid-users@ircache.net
Subject: PATCH: modification to make better use of satellite bandwidth
Resent-Date: Mon, 17 Apr 2000 05:07:06 -0700 (PDT)
Resent-From: squid-users@ircache.net

I have developed some squid modifications based on what I need to make
btter use of a satellite circuit.
I though I'd make them available in case they are of more general
interest. If people show interest in them, I will make the effort to put

them up on a web page with more documentation etc.

We have a satellite circuit with RTT of about 500 millisec. We also have

a terrestrial connection. The satellite is fixed price bandwidth but we
pay by the meg for the terrestrial link.
So what I want to achieve is to use the satellite when possible and fall

over to the the landline when the satellite is overloaded.

The traditional squid setup for this would be to use squids on the other

side of each link as parents and weight them appropropriatly.
The downside with this is that it involves a ICP ping for each request.
The satellite latency is already high and an ICP ping for each request
doubles it.

The solution I have developed is as follows.

A background ping option for parents.
This sends an ICP ping to parents only occasionally (default 30 sec) to
keep the RTT info updated.

A Weighted Round Robin Option for parents.
This divides traffic between parents of this kind proportionally to
their weighted RTT. A parent with half the weighted RTT gets twice the
traffic.

The justification for the weighted round robin option is that using a
standard parent with a background ping option is inappropriate as the
"best" parent would get all the traffic until the next ping was sent.

I also have the problem that the existing weight option for parents
doesnt work well in these circumstances. I would have to set such a high

weight for the satellite that the difference between a slow satellite
link and a fast satellite link gets lost in the noise.
To address this, I added a basetime option for parents. This is
subtracted from the RTT before dividing by the weight when choosing
parents.

Here is the patch for my modifications. The patch is actually against
squid-2.2STABLE5 with the hno patches, but it drops cleanly into
squid-2.3STABLE2.
It seems to be working OK for me. Comments are welcome.
Received on Mon Apr 17 2000 - 06:25:42 MDT