When is HTTP 1

When is HTTP 1.1 not HTTP 1.1?

by Scott Bradner

I maintained in a public forum the other day that HTTP 1.1 was widely deployed because I heard that HTTP 1.1-compliant browser and server software were widely deployed. It turns out that even if my information was right, the implication was wrong.

This year's Association for Comput-ing Machinery (ACM) Special Interest Group on Data Communication (SIGCOMM) meeting was held at Harvard a week ago. It was quite an event, with even more on tap than the usual complement of cutting-edge research papers.

To commemorate the 10th anniversary of the SIGCOMM Lifetime Achievement award, there was an all-day tutorial on the technical history of the Internet, lead by Vint Cerf, that included just about every major player in the technical history of the 'Net. There was an evening panel of all past award winners (other than Jon Postel, who passed away last year). The evening panel was moderated by Ethernet inventor Bob Metcalfe.

During the evening, Metcalfe asked the panel how we could get HTTP 1.1. more widely deployed. This would be a good idea, he said, because use of HTTP 1.1 - last year's revision of the Web transport protocol - might help improve the performance of the Internet. I was sitting in the front row of the audience and made a comment, I thought mostly to myself, that HTTP 1.1 was already widely deployed (implying that it was a silly question.) My comment was picked up by some of the panelists, and Metcalfe moved on to another question.

But the next day I was approached by an AT&T researcher who politely implied that I had no idea what I was talking about. In particular, she had quite a bit of research data that showed that the new functions in HTTP 1.1 that made it better than HTTP 1.0 do not show up very often on the 'Net. Suitably embarrassed, I tried to find out how my information could differ so much from the empirical data she collected.

I made my assertion based on the deployment of software that claims to be compliant to the HTTP 1.1 draft standard, but much of this software turns out to not have the advanced features turned on by default. Because most users would not think of figuring out how to enable these features, they do not get turned on. This means that it's not good enough to get the vendors to implement new features, you also have to get them to turn the features on.

At this point it seems that the claim HTTP 1.1 is widely supported is false advertising. This is not going to make deployment technology upgrades to the 'Net any easier, and it's going to make determining what is going on a lot harder than counting software version numbers.

Disclaimer: Harvard does not need advertising, false or otherwise, and the above is my observation.