An architecture for network resource monitoring in a distributed environment

Abstract

As part of its HiPer-D Program, the United States Navy is developing an experimental distributed system which achieves survivability by dynamically reconfiguring the system using replicated system components and resources. To enable the reconfiguration, resource monitors observe the behavior of the system and report this information to a resource manager. The resource manager makes reconfiguration decisions based on this information. Because all reconfiguration decisions are based on data obtained from resource monitors and the network is the common resource linking all components in the distributed system, this paper focuses specifically on network resource monitoring. A generalized network resource monitor architecture is proposed. Two instantiations of this architecture are then presented. The first is based on custom developed tools tailored to a specific application while the second is based on commercially available products (e.g. SNMP, RMON, etc). Scalability, intrusiveness, and fidelity are identified as evaluation criteria against which implementation trade-offs are made. This paper presents the results of initial experiments as well as future research directions.