Active resource allocation in active networks

A central premise of active networking is that small amounts of user-specif ic computation inserted at key locations within the network can significant ly improve network-based application performance, Hence, a critical issue f or active networking is the amount of resources required to achieve perform ance gains since excessive demands on scarce node resources can have detrim ental effects on the entire network. In this paper, we explore the trade-sp ace between active resource utilization and performance, and attempt to pro vide insights into where limited active resources should be located within a given network topology in order to optimize performance. Simulation exper iments based on performance assessments of the active error recovery/nomine e-based congestion algorithm (AER/NCA) protocol - an active networks-based reliable multicast protocol - reveal that a high percentage of the achievab le performance gains can be attained with only a small number of optimally selected network nodes providing active services. Further experiments show that an early algorithm for dynamically activating-deactivating active serv ices within the network can achieve a significant portion of the gains affo rded by an optimal, static configuration. The implication here is that this or similar algorithms hold significant promise as a means for active netwo rks to dynamically self-optimize active resource allocation.