Varnish is a fun project.

phk has plenty of great ideas, but the Varnish design is tailored towards a specific workload (ie mostly fits in RAM) which handles small objects. Varnish doesn’t take advantage of things that Web Caches should (like object locality on disk) and, well, Squid doesn’t either, but Varnish actively avoids doing so by putting all its faith in the operating system VM system. You can’t scale specific workloads that way.

There’s nothing wrong with the Squid design per se. Plenty of high performance applications look like squid do, they just don’t copy data around as much.

Squid, in its core, is just a big event processing loop. Eventually this’ll have to be modified to support SMP but the basic theory is fine. What we need to do is change how we work with the disk and network to be more efficient and thats precisely what I’m doing.

Ideally Squid in ~ 12 months will look at lot more interesting. Just realise that phk and des are funded to do this and I’m mostly doing this for free.

Squid doesn’t currently support RTSP. We’d love to support it but it’ll take time to code up and none of us actually work for ISPs or companies that have these requirements. We basically need to find funding to code this stuff up.

Send me an email if you’re interested in looking into helping with this.

The Varnish developer claims that the Squid design is something that should be avoided. I would be really interested in hearing your take.