If you Google "AJAX Web 2.0" you'll get over eight million hits, but what technologies will you find in that mix that can truly deliver on the promises of Web 2.0 today? While there's no single definition of Web 2.0, at its heart lays the Internet acting as a platform for social networks, where information can be created and shared in a community of interest. Rich Internet Applications (RIAs) relate to Web 2.0 concepts only in that they enhance the platform by providing a more effective user interface. AJAX relates to Web 2.0 only in that it provides a lightweight approach for developing RIAs that execute through a single ubiquitous interface, the Web browser.

An examination of the interaction models for existing social networking platforms like wikis and blogs reveals that they lack the instantaneous nature of true human interaction. This is an artifact of the synchronous Web model that must be overcome when we consider next-generation platforms envisioned in the Web 2.0 spectrum. Internet-based chat is the most basic example of the near-instantaneous interaction that Web 2.0 demands, but even delivering these basic capabilities in a scalable, lightweight, browser-based mechanism is beyond the scope of most AJAX technologies. In fact, if you sift through those eight million Google hits, you'll find only a handful of open source technologies that address the problems associated with pushing content asynchronously to the user through standard browser mechanisms. We'll provide an overview of those technologies later, but first some basics.

The Basics of Web-based Push
Industry has not standardized on a mechanism or a name for asynchronous data push to the browser. In the AJAX realm the technique goes by a variety of names including AJAX Push, Comet, and Reverse AJAX, but regardless of what you call it, the various approaches share a common characteristic. The only lightweight communication mechanism available to the browser is the HTTP protocol, but HTTP only facilitates standard request/response communication initiated from the browser to the server. To deliver responses asynchronously, it's necessary to invert the HTTP protocol by holding an open request at the server, and fulfilling that request when an update is available. The differences between standard AJAX request processing and the inverted mechanism required for push are illustrated in Figure 1.

To support asynchronous push we need to hold a browser connection open in anticipation of a server-based event that will result in presentation changes in the browser. Intuition suggests that there are issues with this approach and in this case intuition serves us well, since we face connection-related problems at both the browser and the server.

Browser Connection Limit
One of the main difficulties faced by asynchronous Web techniques is caused by the typical browser "two-connection limit." To reduce server load, the HTTP specification recommends that user agents establish at most two connections per server, but asynchronous techniques consume one of these connections as a notification channel. Considering the case where multiple browser windows are open to the same Web page, if each window attempts to establish its own connection, we quickly saturate the available connections. To work within these limitations, a single connection must be shared, but the scripting environments of distinct browser windows are isolated for security reasons, making it difficult to coordinate a shared connection. One technique that can be applied is to make use of a shared HTTP cookie, which is associated with all HTTP interaction with a given server. The cookie is shared across windows and, by manipulation through JavaScript, can be used for cross-window communication.

The problem is amplified in portal environments because legacy portal engines don't support the necessary asynchronous communication path. For push to work in multiple portlets it's not only necessary to share the browser connection, but a single shared server connection is required to marshal asynchronous responses. If portlets are deployed from multiple WARs, there's an additional complication in that the sharing mechanism must coordinate between multiple Web applications and will require some IPC mechanism to do so.