Distributed Policy Management and Comprehension with Classified Advertisements

Abstract

Distributed systems present new challenges to resource management, which cannot be met by conventional systems that employ relatively static resource models and centralized allocators. Matchmaking paradigms based on identifying compatible classified advertisements (Class-Ads) placed by providers and requesters of services work well in such environments. Condor is a production-quality distributed system with a distributed policy model that uses ClassAds for matchmaking. However, due to the distributed polices and dynamics of such environments, understanding why some ClassAds are not matched while others are can be a very complex task. We therefore present algorithms which not only identify problematic aspects of a policy, but also suggest modifications.