Exploiting Application Tunability for Efficient, Predictable Parallel Resource Management
Abstract
Parallel computing is becoming increasingly central and mainstream driven both by the widespread availability of commodity SMP and high-performance cluster platforms as well as the growing use of parallelism in general-purpose applications such as image recognition virtual reality and media processing. In addition to performance requirements, the latter computations impose soft real-time constraints. necessitating efficient, predictable parallel resource management. Unfortunately, traditional resource management approaches in both parallel and real-time systems are inadequate for meeting this objective; the parallel approaches focus primarily on improving application performance and/or system utilization at the cost of arbitrarily delaying a given application. while the real-time approaches are overly conservative sacrificing system utilization in order to meet application deadlines. In this paper we propose a novel approach for increasing parallel system utilization while meeting application soft real-time deadlines. Our approach exploits the application tunability found in several general-purpose computations. Tunability refers to an application's ability to trade off resource requirements over time, while maintaining a desired level of output quality. In other words, a large allocation of resources in one stage of the computation's lifetime may compensate. in a parameterizable manner, for a smaller allocation in another stage. We first describe language extensions to support tunability in the Calypso programming system, a component of the MILAN metacomputing project, and evaluate their expressiveness using an image processing application. We then characterize the performance benefits of tunability, using a synthetic task system to systematically identify its benefits and shortcomings.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1998
- Accession Number
- ADA439729
Entities
People
- Fangzhe Chang
- Vijay Karamcheti
- Zvi Kedem
Organizations
- New York University