W3CArchitecture DomainActivity Statement

HTTP
Activity Statement

Work on the HTTP Activity has been managed as part of W3C's Architecture Domain. This Activity has achieved its goals, and has been closed in May 2000.

  1. Introduction
  2. Role of W3C
  3. Current Situation
  4. What the Future Holds
  5. Contact

Introduction

As Web traffic has come to dominate the Internet, remedying the weaknesses of the HTTP protocol has become critical. W3C has worked with the Internet Engineering Task Force (IETF) to develop a number of refinements to HTTP, work that has culminated in a new specification for the protocol. This new specification, called HTTP/1.1, has just recently become an IETF Draft Standard.

IETF specifications proceed through three main steps: Proposed Standard; Draft Standard; and finally Internet Standard. A Draft Standard is considered to be the final specification, apart from only small changes of a very specific nature. Only a short step away is the Internet Standard - a specification that, to quote the IETF, is "stable and well-understood, is technically competent, has multiple, independent, and interoperable implementations with substantial operational experience, enjoys significant public support, and is recognizably useful in some or all parts of the Internet."

The standardization of HTTP was started within the IETF in late 1994, and W3C has strongly supported these efforts. Significant W3C effort went into HTTP/1.1, particularly from Jim Gettys, visiting scientist at W3C from Compaq, HTTP/1.1 editor and co-author, and from Henrik Frystyk Nielsen, formerly W3C staff, and co-author of HTTP/1.1.

W3C has several HTTP/1.1 implementations and a large community of developers have already demonstrated their interest. Experimental implementations include the libwww client library, and Jigsaw, W3C's Web server. These were among the very first HTTP/1.1 implementations, and they played a key role in discovering errors in the HTTP/1.1 Proposed Standard, and show the full potential of the protocol.

Features of HTTP/1.1

The focus of the work behind the HTTP/1.1 specification has been to alleviate the most prominent problems in HTTP/1.0 which has led to serious bottlenecks on the Internet as the Web continues to grow. The result has been the specification of a protocol which will make the Web faster and more efficient. The three main features of interest are as follows:

1) Support for Virtual Hosting

The rapid growth of the Web has produced a frenzy for domain names like mycompany.example.com, often as important for corporate recognition as a logo. Domain names may be infinite in number, but the IP addresses they translate into are not, and IP address depletion has become a serious concern. The HTTP/1.1 Host header field allows Web service providers to assign multiple domain names to a single IP address in such a way that a Web server can distinguish the home page for mycompany.example.com from yourcompany.example.net without using more than one IP address.

2) Requests for information handled more efficiently

HTTP uses the Internet TCP/IP protocol stack. All information you read or write on the Web is sent accross the Net in TCP/IP packets. A TCP connection is really like a responsible courier getting around in a big city (the Net) - it makes sure the data you send and receive reaches the final destination realiably while avoiding traffic jams and allowing other people to get through as well. The funny thing is that TCP drives an old car - it takes time for it to warm up, and as soon as it is done, it cools off again very quickly.

To function efficiently, HTTP must take advantage of TCP/IP's strengths and avoid its weaknesses, something that HTTP/1.0 does not do very well. Whenever a client accesses a document, an image, a sound bite etc. HTTP/1.0 creates a new TCP connection and as soon as it is done, it is immediately dismissed and never reused. As a result, TCP rarely has time to get warm leaving lots of "cold cars" with little data creating a lot of traffic jams.

HTTP/1.1 fixes this in two ways. First, it allows the client to reuse the same TCP connection (persistent connections) again and again when talking to the same server. Second, it makes sure that the courier carries as much information as possible (pipelining) so that it doesn't have to run back and forth as much. That is, not only does HTTP/1.1 use less TCP connections, it also makes sure that they are better used. The result is less traffic jam and faster delivery.

3) Efficient Caching

Documents you read on the Web are often read by thousands and even millions of other people at the same time. This of course keeps servers very busy. Imagine that instead of having everybody talking to the same server people could get the same information much closer to where they are. This is what caching allows us to do.

Whereas HTTP/1.0 merely enabled caching, it did not specify any well-defined rules describing how a cache should interact with clients or with origin servers. The lack of control resulted in that most content providers and users did not trust the HTTP/1.0 caching model and instead tried to short-circuit it. The result was that many busy parts of the Internet were bogged down even more. A major part of the HTTP/1.1 specification is devoted to providing a well-defined caching model which allows both servers and clients to control the level of cachability and the conditions under which the cache should update its contents.

Digest Authentication

Another important part of HTTP/1.1 is the Digest Authentication Specification. Digest authentication allows users to authenticate themselves to a server without sending their passwords in clear text which can be sniffed by anybody listening on the network. In HTTP/1.0, passwords are sent without being encrypted using so-called basic authentication. Although not providing real security, Digest Authentication is an important step in a making the Web a more secure place to live.

The HTTP Extension Framework

A continuing area of interest is how HTTP can be extended according to the needs of specific applications. HTTP has been extended locally, as well as globally, in ways that few could have predicted. Current efforts span an enormous range, including distributed authoring, collaboration, printing, and remote procedure call mechanisms. The usual practice is to add new header fields to the protocol, and rely on the software at the other end to recognize the header and process it accordingly. This, however, is the equivalent of relying on magic! A standard framework for defining extensions has for some time been badly needed.

The HTTP Extension Framework provides a simple yet powerful mechanism for extending HTTP. The Framework enables authors to introduce extensions in a systematic manner: programmers will be able to specify which extensions are introduced along with information about who the recipient is, and how the recipient should deal with them.

The specification has been accepted as an experimental RFC as of February 2000 (RFC 2774). Here are some W3C specifications that are already using it:

Composite Capability/Preference Profiles (CC/PP)

The Composite Capability/Preference Profiles (CC/PP) is a mechanism for describing the capabilities and preferences associated with users and the hardware and software they are using to access the World Wide Web.

Platform for Privacy Preferences (P3P)

The P3P specification will enable Web sites to express their privacy practices and users to to exercise preferences over those practices. P3P products will allow users to be informed of site practices, to delegate decisions to their computer when possible, and allow users to tailor their relationship to specific sites.

Jigsaw has already an experimental implementation of the extension framework and implementation is ongoing in the libwww.

Role of W3C

The HTTP drafts and specifications are produced by the IETF HTTP Working Group, in which the W3C Team takes an active role in editing specifications and providing sample implementations. Related protocols include the HTTP-NG specifications and the HTTP Extension Framework.

Current Situation

Libwww and HTTP/1.1

Libwww is a highly modular, general-purpose client-side Web API written in C for Unix and Windows. It is well suited for both small and large applications, such as browser/editors, robots, batch tools, etc. Pluggable modules provided with libwww include complete HTTP/1.1 (with caching, pipelining, PUT, POST, Digest Authentication, deflate, etc). As for all W3C OpenSource code, the purpose of libwww is to provide an environment for experimenting with extensions and new features. The focus of libwww is performance, modularity, and extensibility. Libwww now supports a large community of authors and boasts a number of new features: HTML 4, XML, RDF, SSL and much more.

Jigsaw and HTTP/1.1

Jigsaw is not only a server, it also provide a reusable HTTP/1.1 stack, with a RFC 2616 compliant cache in its latest 2.1.1 release. It allows every Java program to use a HTTP/1.1 compliant stack without modification.

The ongoing results from our analysis show that the performance is increased significantly using HTTP/1.1 with persistent connections and pipelining. We also have practical experience with clients such as Amaya, who use the pipeline features of libwww.

HTTP Extension framework

As of February 2000, HTTP Extension Framework is now known as RFC 2774.

What the Future Holds

No more developments are scheduled, now that HTTP Extension Framework has been accepted as a RFC. However we will continue to survey all the ongoing activities around HTTP/1.1

Contact

Yves Lafon, W3C