# pro-cow : protocol compliance on the web

From: Balachander Krishnamurthy <bala@research.att.com>
Date: Tue, 03 Aug 1999 14:52:07 -0400
Message-Id: <199908031852.OAA56918@raptor.research.att.com>


[abstract, observations, and url of a paper that might be of interest.
apologies for those of you who get duplicate copies]

abstract

With the recent (draft) standardization of the HTTP/1.1 protocol on
the Web, it is natural to ask what percentage of popular Web sites
speak HTTP/1.1 and how {\it compliant} are these so-called HTTP/1.1
servers.  We attempt to answer these questions through a series of
experiments based on the protocol standard. The tests are run on a
comprehensive list of popular Web sites to which a good fraction of
the Web traffic is directed. Our experiments were conducted on a
global extensible testing infrastructure that we built to answer the
above questions.  The same infrastructure will be used to answer
questions like the percentage of the traffic flow that is end-to-end
HTTP/1.1.  Our results show reasons for concern on the state of
HTTP/1.1 protocol compliancy and the subset of features actually
available to end users.

some interesting observations:

we tested over 500 'popular' soi disant HTTP/1.1 sites and only about 60%
were unconditionally compliant on 3 basic tests (GET, HEAD, dealing with
absence of Host header). 7% failed all three tests. 70% of the sites handled
persistent connections and nearly that many pipelining. Half of the sites
handled range requests. 20% of sites didn't support any of persistent
connections/pipelining/range. 30% of sites passed all 6 tests.

if u want to look at the paper (jointly written with martin arlitt of hp-labs)