An adaptive, distributed algorithm for interest management

Abstract

The scale of large-scale virtual environments (LSVEs) is limited by the ability of the supporting infrastructure to deliver data to participants in a timely manner. Multicast can improve data delivery time by minimizing message send time similarly to broadcast while reducing the delivery of extraneous messages which goes with broadcast. However, multicast groups are typically limited resources, mostly due to hardware limitations. Significant performance improvements have been made using judicious, static assignments of multicast groups based on pre-defined criteria such as geographic location. However, such static approaches ultimately lack the flexibility to scale to meet the requirements of highly dynamic LSVEs. Dynamic multicast grouping has been considered to be too computationally expensive to be practically applicable. This dissertation derives a straightforward heuristic based on readily available data from which various computationally inexpensive algorithms can be derived. Through experiments with simulations and a well-known LSVE environment, the feasibility of general application of these algorithms is demonstrated, as well as the significant reduction in the use of multicast groups they achieve. Finally, experimentation and analysis demonstrate that the real issue with dynamic assignment of multicast groups is the time required to reconfigure multicast hardware.