I'm glad to see Yahoo put this project out in the open source world. What I'm having a hard time with is understanding precisely where it fits after having read through all the public docs. Despite the comparisons drawn on the intro page, Traffic Server seems close to squid and varnish functionally, but really only close to squid in actual implementation: a highly configurable split ram/disk caching proxy. Varnish is quite a departure from this approach (http://varnish.projects.linpro.no/wiki/ArchitectNotes), and nginx at its core is not a caching proxy at all. This doesn't diminish its value as a system that's served tremendous traffic volumes over the last decade, but I'd love a more complete description of why I should choose TS over the alternatives in caching and reverse proxy roles.
Ha, I was just googling a range of years along with keywords trying to find that very article.
Traffic Server looks to be well designed and implemented (especially compared to Squid), but it's still based on exactly the same idiotic 1975 assumptions.
It's been in for about five months and seems to be undergoing heavy development. Like everything else nginx I'm sure it will be rock solid, but I don't think it can be considered core yet.
Note that Squid is also heavily used in Yahoo's internal caching system. I suppose there must be reasons for that.
But, this is certainly a good thing. Traffic Server was an extremely good product in its heyday (my previous Squid-based company competed with Inktomi-based products), and maybe an Open Source version will be good for the eco-system in general. Varnish gets mentioned a lot as a super fast alternative to Squid, but it has a very limited set of use cases compared to Squid.
A fine pocketknife has a very limited set of use cases compared to a Leatherman.
It also is actually good at a limited set of tasks, instead of just being mediocre-to-terrible at %90 of everything some bureaucrat thought might be useful 5+ years ago, all in a monolithic pile with awful configuration.
And, there are no bureaucrats involved in the development of Squid. The Squid team are old friends of mine, and they are smart, competent developers working with a much larger set of problems and compatibility issues in a much larger codebase than most of us have to contend with, and they do a fine job of it. As is shown by the market leading position of Squid; it is the most popular Open Source web caching proxy server, by far.
How does it work for Comet style apps? I am not sure, but I think one would need a proxy server that can keep a lot of connections open. Not sure if classical proxy servers are designed to do that?