I thought I was joking... [apologies to non-geeks]

[blufive]

A few days ago, in a web authoring forum, someone suggested to me that HTTP was just a guideline, and that implementors could pick and choose which bits they wanted to implement. I responded (rather more testily than necessary, but the guy was being an arrogant smartass) that maybe we should play fast-and-loose with TCP/IP while we were at it.

Well, someone's done some poking around and discovered that Internet Explorer and IIS do just that. So Microsoft are buggering about with the protocol that runs the Internet in order to make their web server and browser seem faster. Lovely.

[later: hey! I scooped slashdot! The downside to this is that the site is now slashdotted. The poor webserver is now a gibbering wreck. Hopefully, you'll be able to see the page again in a few days...]

[even later: following a prod from sbisson, I've done a bit of research, and I now suspect that the original article linked above contains insufficient information to determine if MS is doing something wrong - the descibed behaviour MAY be valid in certain circumstances, but we need to know more about the test conditions to be sure. And I _SO_ wanted to spend my Sunday evening resarching TCP/IP and HTTP 1.1]

[much later: after more research, I'm fairly sure that IE is behaving, and the original post is wrong. Guess I should crank up my "scepticism" setting.]

[link: mpt]

Comments

[sbisson.livejournal.com]

Err, surely that's just HTTP 1.1 in action. IIS is by default a HTTP 1.1 server, and that just looks like perfectly normally HTTP 1.1 behaviour to me...

HTTP 1.1 is designed to leave connections open for browsers to access pages in just the described manner...

[blufive.livejournal.com]

We're heading a bit beyond my expertise here. He's talking about TCP/IP, rather than HTTP [goes and reads bits of http1.1 spec] but you may be right.

There's some debate on the corresponding slashdot thread about this...

[blufive.livejournal.com]

(oops, that was supposed to be a reply to sbisson's post above. oh well)

The consensus over on slashdot is that (a) someone ought to do something about slashdotting poor innocent webservers, (b) IE sucks, use mozilla/Konquerer/etc. instead, and (c) Microsoft own the world, stop whining already.

Well there's a surprise.

Between all the usual crud, however, are some people suggesting it's just HTTP 1.1 in action, and a few more saying oh no it isn't. Some are saying well, it is, sorta...

[karmicnull.livejournal.com]

IE uses the standard Windows NDIS IP stack. HTTP as a protocol sits above TCP/IP. The fact that IE/IIS chooses to leave 1/2 the connection open upon download completion rather than closing the TCP port full duplex is an HTTP issue rather than an IP one. Whether it is spec-compliant or not I don't know, as I'm no HTTP expert.

[blufive.livejournal.com]

Thanks for that. I'm still researching to try and figure out myself what's going on here. I'm now pretty sure that the original report is seriously lacking relevant details, so in my book, IE gets the benefit of the doubt for now...