Lab Home Lab Phone Lab Search
Home Research Papers Software People Jobs Los Alamos Photos Site Map

Rate Adjusting Protocol for Internet Delivery

As server infrastructures evolve to support services such as video-on-demand, web caching and hosting, and remote scientific visualization, the ability of the Internet to deliver such services in light of an exponentially increasing volume of traffic becomes increasingly difficult. This is particularly true of (1) Grand Challenge applications that require interactive and remote computational steering of large visualization data sets over the wide-area network (WAN) and (2) multimedia applications such as RealPlayer, Quicktime Pro and Windows Media Player.

In the case of many multimedia applications, their unresponsiveness to network congestion places unfair demands on the network. These applications generally blast UDP packets across a network regardless of network congestion. Consequently, network congestion increases and the performance of applications which use TCP, a protocol that is considered to be a "good network citizen" because it reduces its transmission rate in the face of network congestion, is dramatically reduced. All because UDP, unlike TCP, has no self-regulating mechanism.

Active queue-management schemes for routers are being proposed to punish these non-adaptive applications by dropping packets from their flows to ensure that well-behaved TCP applications do not starve for network resources. As a result, however, the performance of multimedia applications will be crippled.

To maintain the performance of such applications while making them "good network citizen", we propose the Rate Adjusting Protocol for Internet Delivery (RAPID) which is an adaptive, rate-based, TCP-friendly network protocol. RAPID can be used by multimedia applications to achieve good performance in the presence of active queue management without penalizing other applications.

RAPID also allows congestion control to be disabled, thus turning RAPID into a reliable UDP implementation that is appropriate for use in dedicated network environments, e.g., direct bulk-data transfer of visualization data sets from a compute engine to a visualization engine.

Publication List