The project is about a year old.
Squid was part of the Harvest project and was split when commercial Harvest came into existence. The current version of Squid is 1.1, but 2.0 should be available later this year.
Squid requires a lot of physical memory (128 to 256M) and the disk is the bottleneck. There is little support for HTTP/1.1 features like persistent connections or pipelining.
The Internet Cache Protocol is now being standardized in IETF. Currently, ICP is being used for passing messages, but should really be used to enforce policy. ICP supports multicast.
Copyright: Is the real issue revenue or the law? Will hit-metering help this problem?
Hit-Metering: An Internet draft has been submitted.
Cache-busting: Some servers are preset to defeat this. Some cache operators violate HTTP to force documents to cache even though they are not supposed to be cached.
Configuration is complex: How can that be done and still maintain tight access controls? Refresh parameters need to be similar amongst all members of the group.
Optimal Cache Forwarding: The goal should be to forward requests toward origin. This can be done using ICMP queries.
Transparent caching: outbound TCP has to be intercepted.
A possible Goal for a Global Cache Mesh: Each AS runs a cache that has a well-formed name. Another possibility is to have each AS have a multicast group.
For more information, see http://www.nlanr.net/Cache and http://squid.nlanr.net
One individual suggested the use of "Sonar" to do the ICMP redirects.
Steve Corbato asks about the types of organizations participating:
ISPs(50%), Educational Institutions(30%), Corporate entities(20%)
John Stewart: Do you have any feedback from users on performance?
Not to much. A three level hierarchy is about all users will stand.
Dorn of Epoch Networks notes that there is a new NAP in Atlanta http://www.epoch.com/aix