Abstract

Although a large number of Web applications are based on the publish/subscribe paradigm, including e.g. RSS and Twitter feeds, the underlying architecture is based on the client/server paradigm. The HTTP long-polling mechanism is a commonly used approach to overcome the situation; the clients simply leave an open HTTP GET pending at the server. Whenever an update becomes available, the server pushes the new data separately to each of the clients over the open connections, after which the clients need to post another HTTP GET request to receive further updates. This incompatibility between the application needs and the underlying architectural approach leads to a number of inefficiencies, including the extra TCP state at the servers for the open long-poll connections, the clients needing to constantly re-subscribing to updates, and the inability to use multicast.

In this presentation, we explore a couple of opportunities opened up for event and content delivery by employing a native publish/subscribe approach. In other words, we explore a number of options on how to use the PSIRP native pub/sub core, including native multicast and in-network caching, to improve the overall efficiency of typical Web applications. In particular, we briefly introduce the idea on how to extend a single TCP connection to cover multiple parallel receivers with the help of overlay multicast and in-network caching, and then describe another idea of how to proxy the WebSockets and Google PubSubHubbub (PuSH) with the native in-network caching PSIRP pub/sub.

Homepage: http://www.tml.tkk.fi/~pnr/