HPC Application Energy Profiling for Energy Optimization
Energy consumption is quickly becoming one of the primary bottlenecks of compute clusters and supercomputers. The DOE is a primary developer and consumer of these power hungry supercomputers, however smaller cluster machines are also widely used through other government agencies and industries. Yet, there are few power profiling tools that allow application developers transparent insight into the power requirements of their application code. Software tools and APIs will be developed to allow application developers to gather power profiling data from existing power sensors (e.g., IPMI) and to correlate the power information with application methods and kernels. To compliment the power profiling software, additional node sensors will be developed to increase the transparency into the component (e.g., CPU, GPU, and DRAM) power usage. Commercial Applications and Other Benefits: The power reduction benefits would benefit developers and end-users of supercomputing systems in both government agencies and industry. The developers would benefit from reducing application power consumption via the use of the proposed tools, thus increasing the value of their application to end users resulting in a competitive advantage. End users would benefit from the reduced energy costs, reduced building requirements, and the socio-political advantage of touting reduced energy consumption.
Small Business Information at Submission:
Research Institution Information:
Rnet Technologies, Inc.
240 West Elmwood Drive Suite 2010 Dayton, OH 45459
Number of Employees:
The Ohio State University
1960 Kenny Road
Columbus, OH 43210