Performance and Energy Management in High Performance Computing Systems Using Application-Level Behavioral Attribute Driven Techniques
Small Business Information
Pc Krause And Associates, Inc.
3000 Kent Avenue, Suite C1-100, West Lafayette, IN, 47906
AbstractThe objective of the proposed effort is to characterize High Performance Computing (HPC) subsystem interactions into meaningful metrics and correlates and, determine the correlations between high-level application behavioral metrics to power consumption in HPC systems. The proposed effort will also evaluate the potential effectiveness of using the correlates as inputs to resource management and job scheduling algorithms for the purpose of consistent energy efficient HPC system operation. A combination of experimental and simulation-based approaches will used to conduct a series of HPC system performance evaluations to further identify HPC sub-system interactions, their conversion into application behavioral metrics and their potential as correlates. Correlation studies on application behavioral metrics will be performed to establish power consumption saving potential. Effectiveness of the correlates for system energy conservation will be evaluated by including the correlates as simulated job scheduler inputs. Commercial Applications and Other Benefits: The main commercial opportunity made available by this effort will be through additional functionality and system features that would enhance current simulation products. A power aware scheduling capability will provide a resource management model that is scalable and supports the inter-connection of disparate software applications running concurrently on a cluster of multi-core computing nodes. The resulting product will be widely applicable to both military and commercial computing applications that seek run time and power efficient execution among disparate applications. The inherent compatibility with multi-core and cluster architectures will enable rapid proliferation as the adoption of these novel architectures by industry continues to increase thereby further driving down the cost of computing. Sales of intelligent HPC energy load management control systems designed under this effort could provide additional commercial opportunities.
* information listed above is at the time of submission.