Deadline-sensitive workflow orchestration without explicit resource control
Journal of Parallel and Distributed Computing
| Hi-index | 0.00 |
Advances in server virtualization offer new mechanisms to provide resource management for shared server infrastructures. Resource sharing requires coordination across self-interested system participants (e.g., providers from different administrative domains or third-party brokering intermediaries). Assignments of the shared infrastructure must be fluid and adaptive to meet the dynamic demands of clients. This thesis addresses the hypothesis that a new, foundational layer for virtual computing is sufficiently powerful to support a diversity of resource management needs in a general and uniform manner. Incorporating resource management at a lower virtual computing layer provides the ability to dynamically share server infrastructure between multiple hosted software environments (e.g., grid computing middleware and job execution systems). Resource assignments within the virtual layer occur through a lease abstraction, and extensible policy modules define management functions. This research makes the following contributions: (1) Defines the foundation for resource management in a virtual computing layer. Defines protocols and extensible interfaces for formulating resource contracts between system participants. Separates resource management functionalities across infrastructure providers, application controllers, and brokering intermediaries, and explores the implications and limitations of this structure. (2) Demonstrates policy extensibility by implementing a virtual computing layer prototype, Shirako, and evaluating a range of resource arbitration policies for various objectives. Provides results with proportional share, priority, worst-fit, and multi-dimensional resource slivering. (3) Defines a proportional share policy, WINKS, that integrates a fair queuing algorithm with a calendar scheduler. Provides a comprehensive set of features and extensions for virtual computing systems (e.g., requests for multiple resources, advance reservations, multi-dimensional allocation, and dynamic resource pools). Shows the policy preserves fairness properties across queue transformations and calendar operations needed to implement these extensions. (4) Explores at what layer, and at what granularity, decisions about resource control should occur. Shows that resource management at a lower layer can expose dynamic resource control to hosted middleware, at a modest cost in fidelity to the goals of the policy.