Utilization-based techniques for statically mapping heterogeneous applications onto the HiP.pdf

UTILIZATION-BASED TECHNIQUES FOR STATICALLY MAPPING HETEROGENEOUS APPLICATIONS ONTO THE HIPER-D HETEROGENEOUS COMPUTING SYSTEM? SHOUKAT ALI? , JONG-KOOK KIM? , YANG YU? , SHRIRAM B. GUNDALA? , SETHAVIDH GERTPHOL? , HOWARD JAY SIEGEL§?, ANTHONY A. MACIEJEWSKI§ , AND VIKTOR PRASANNA? Abstract. This research investigates the problem of allocating a set of heterogeneous applica- tions to a set of heterogeneous machines connected together by a high-speed network. The proposed resource allocation heuristics were implemented on the High Performance Distributed Computing Program’s (HiPer-D) Naval Surface Warfare Center testbed. The goal of this study is to design static resource allocation heuristics that balance the utilization of the computation and network re- sources while ensuring very low failure rates. A failure occurs if no allocation is found that allows the system to meet its resource and quality of service constraints. The broader goal is to determine an initial resource allocation that maximizes the time before run-time re-allocation is required for managing an increased workload. This study proposes two heuristics that perform well with respect to the load-balancing and failure rates. These heuristics are, therefore, very desirable for HiPer-D like systems where low failure rates can be a critical requirement. 1. Introduction. With the widespread use of increasingly powerful commer- cial off-the-shelf (COTS) products, some system designers have started a shift from custom development to COTS-based systems to achieve lower costs and more flexi- ble systems [22, 26]. However, to use COTS-based systems effectively as parts of a larger system, one needs to exploit the heterogeneity in processor speeds, memory structures, specialized hardware capabilities, etc., that most likely will be present in different COTS products. Heterogeneous computing (HC) is the coordinated use of different types of machines, networks, and interfaces to meet the requirements of widely va