Re: 1.2 beta 18 success and a hack

Martin Hamilton writes:

>-----BEGIN PGP SIGNED MESSAGE-----
>
>Hi, just to report that we've had 1.2 beta 18 up and running and being
>torture tested (with lots of {tcp,udp}-banger processes running on
>URLs from old cache logs) on one of our RedHat 5 boxes for some ten
>days now (without a single crash!) and today felt brave enough about
>it to run it live on one of our production service machines. Looking
>good so far :-)
>
>I'd also like to point people at
>
> <URL:http://wwwcache.ja.net/~martin/no-misses.diff>
>
>This is a hack to 1.2 beta 18 which adds support for a new option in
>squid.conf - "no-misses" in a cache_peer entry. When this is enabled
>for a peer, the machine making the ICP request will set a bit in the
>ICP header which is meant to be interpreted by the machine on the
>receiving end as an indication that the requesting machine would
>prefer not to receive an ICP_MISS reply to this request.
>
>It's just a first stab, but seems to work OK between my two test
>machines (SPARC 20 with Solaris 2.6 and ThinkPad 560 with Linux
>2.0.33/2.1.91). Next step is to add some tweaks to Squid's timing so
>that the lack of ICP_MISSes doesn't slow everything down drastically ?
>
>What do people think ? The ultimate aim of this for me is to be able
>to multicast the ICP requests out, and only get replies back from
>those peers which have the cached URL. Each peer still has to do work
>in order to figure out whether they have the requested URL, but they
>don't all have to do the work of replying to each ICP request they
>receive.

Sounds very cool.

I've got lots of changes for beta19 which I hope to have out tonight.

Duane W.
Received on Tue Jul 29 2003 - 13:15:47 MDT