|
Abstract : |
Federated distributed systems present new challenges to resource management, which cannot be met by conventional systems that employ relatively static resource models and centralized allocators. We previously argued that Matchmaking provides an elegant and robust resource management solution for these highly dynamic environments [5]. Although powerful and flexible, multiparty policies (e.g., co-allocation) cannot be accomodated by Matchmaking. In this paper we present Gang-Matching, a multilateral matchmaking formalism to address this deficiency. 1. Matchmaking Resource management via Matchmaking occurs as a four-step process. Entities (i.e., servers and customers) requiring matchmaking services express their characteristics, constraints and preferences to a Matchmaker in classified advertisements (Step 1). Figure 1 shows a workstation's classad, where the Constraint and Rank expressions identify the machine's constraints and preferences respectively. Attributes of candidate classads are accessed via the pseudo-attribute other. The Matchmaker employs a very generic matchmaking algorithm to create matches incorporating the constraints and preferences of entities (Step 2). Matched entities are then notified, and their classads discarded (Step 3). Finally, matched entities establish an allocation through a claiming process that does not involve the Matchmaker (Step 4). Many complex and useful policies may be defined within this framework; interested readers are referred to reference [6] for sophisticated real-world examples. 2. The License Management Problem Many programs employ software licenses to implement program use policies. For example, the program may be valid only to certain users, or on certain workstations or subnets. Software licenses must therefore be managed as first-, |
