A Policy Language for Distributed Resource Management

互联网提供了有效和公平分配分布式计算资源的潜力，例如高速计算机和数据库，给那些最能有效使用它们的人。实现这一潜力需要轮流描述资源的生产者和消费者，以及在双边约束下分配资源的机制：资源的消费者可以规定资源的可接受标准，资源的提供者可以对消费者施加限制。分类广告语言为定义这些约束提供了精确而强大的表示法，配对算法提供了强制执行这些约束的机制。本研究通过发现和利用分类广告集的同质性来提高匹配效率。它还将配对范例扩展到“共同匹配”，以支持这样的情况：成功的分配需要集合完整和一致的资源集，每个资源集都可能对集合中的其他资源施加限制。这三种增强是由Condor分布式资源管理系统中出现的实际问题所驱动的，该系统被计算科学家广泛用于利用空闲工作站的能力。这项研究的结果将通过直接应用它来提高秃鹰的服务来检验。

Marvin H. SolomonCCR-9972232The internet presents the potential for efficient and equitable allocation of distributed computing resources, such as high-speed computers and databases, to those best able to use them productively. Realization of this potential requires a rotation for describing producers and consumers of resources and a mechanism for allocating them subject to bilateral constraints: Consumers of resources may specify acceptability criteria for resources, and purveyors of resources may impose restrictions on consumers. The Classified Advertisement language provides a precise and powerful notation for defining these constraints and the Matchmaking algorithm provides a mechanism for enforcing them. This research improves the efficiency of Matchmaking by discovering and exploiting homogeneous properties of sets of Classified Advertisements. It also extends the Matchmaking paradigm to "co-matching" to support situations in which a successful allocation required assembling a complete and consistent set of resources, each of which may impose restrictions on others in the set. Thee enhancements are motivated by practical problems arising in the context of the Condor distributed resource management system, which is widely used by computational scientists to harness the power of idle workstations. The results of this research will be tested by applying it directly to improve the Condor service.