Thanks for the comments, Cliff. On Nov 12, 11:46 am, Cliff Spradlin <[email protected]> wrote: > Hi, > > I like some of the ideas behind SPDY, but I find its implementation to > be too complex vs. its rewards.
Our goal is to make it so that the speed improvements are compelling enough to merit the additional complexity. > Out-of-order pipelined responses can be done by extending the HTTP > spec to include a 'request-id' or something along these lines. If you > have a long-running stream of incoming, it's okay for that stream to > be created using a new socket.. There are a number of issues here. You're basically talking about doing the framing as part of the HTTP header; I'd be happy to have a performance bake-off between SPDY and an implementation like this. I'm fairly confident that an approach such as this would have minimal performance benefits; but I am happy to be proven wrong with data. > I think the SCTP-like protocol on top of TCP destroys the inherent > debuggability of HTTP and the internet message format. It is harder to debug, that is true. > Along the same lines.. I feel like instead of the custom header > compression being done here, you could do one of two things. Either > use short-form headers (the ones specified in the IMF like 'l' for > Content-Length'), or just compress the entire stream as a single gzip > instance. If the spec is going to send the data by default as gzip > anyway, might as well compress the headers as part of the same gzip > instance. We just compress the headers with gzip, so maybe that is what you're asking for already? I believe more research can be done to make it better. We took a very simple approach here. > I think the use-case of switching tabs to be a reason to implement > priority (and therefore SCTP-like streams on TCP) to be pretty weak.. > people in multiple tabs generally are looking at different sites, and > so changing the priority for the connection in the background tab > isn't going to make any difference. We all have anecdotal ideas, but I can tell you that browsers do struggle with this issue. When we implement it, we'll measure the gains and publish how well it works. If you're right, and it doesn't work, then we won't propose it :-) But I can tell you for certain that browsers struggle with this issue today. Ever open multiple pages from the NYTimes and expect them to load? Which should load first? How can the browser help the server make sure that the foreground tab loads first? > > Basically what I'm saying is.. please implement SPDY with only minimal > modifications to HTTP. I think that with some minimal modifications > you can get the majority of the performance increase you're looking > for. Our goal is speed. I hope others will find the benefits compelling enough to take the (hopefully minimal) complexity. We're also trying to make it so that application-level http semantics don't change. Most application writers shouldn't notice. Mike > > P.S. I think the idea of being able to do an IP redirection without > changing hostname is extremely cool. -- Chromium Discussion mailing list: [email protected] View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-discuss
