Evaluation of a Resource Selection Mechanism For Complex Network Services

Providing complex (resource-intensive) network services is challenging because the resources they need and the re-sources that are available can vary significantly from re-quest to request. To address this issue, we have proposed a flexible mechanism, called active frames, that provides a basis for selecting a set of available distributed computing resources, and then mapping tasks onto those resources. As a proof of concept, we have used active frames to build a remote visualization service, called Dv, that allows users to visualize the contents of scientific datasets stored at remote locations. In this paper, we evaluate the performance of ac-tive frames, in the context of Dv. In particular, we address the following two questions: (1) What performance penalty do we pay for the flexibility of the active frames mecha-nism? (2) Can the throughput of a service based on active frames be predicted with reasonable accuracy from micro-benchmarks? The results of the evaluation suggest that the overhead imposed by active frames is reasonable (roughly 5%), and that simple models based on micro-benchmarks can conservatively predict measured throughput with rea-sonable accuracy (at most 20%).