Runtime power monitoring in high-end processors: methodology and empirical data
Canturk Isci, Margaret Martonosi
Abstract
Canturk Isci, Margaret Martonosi
Abstract
Power modeling and measurement are crucial elements for power aware research and there have been several attempts at varying levels of detail to address good and yet efficient modeling and measurement methodologies. In our conference paper we introduced our live, runtime power estimation project and presented a synchronized real power measurement and power modeling technique at runtime for Intel Netburst architecture, P4-Willamette core implementation, using performance counters as a means to estimate component access rates. We used this access information for component power approximation. This report provides the complete set of used access rate heuristics and expands the results presented in the conference paper to a larger group of benchmark applications. Overall, we conclude that performance counter based component power estimation provides useful results for runtime power modeling and program power phase analysis regardless of any apriori program structure knowledge. 1
OpenAlex reports 495 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Power modeling and measurement are crucial elements for power aware research and there have been several attempts at varying levels of detail to address good and yet efficient modeling and measurement methodologies. In our conference paper we introduced our live, runtime power estimation project and presented a synchronized real power measurement and power modeling technique at runtime for Intel Netburst architecture, P4-Willamette core implementation, using performance counters as a means to estimate component access rates. We used this access information for component power approximation. This report provides the complete set of used access rate heuristics and expands the results presented in the conference paper to a larger group of benchmark applications. Overall, we conclude that performance counter based component power estimation provides useful results for runtime power modeling and program power phase analysis regardless of any apriori program structure knowledge. 1
Key concepts: Computer science, Power (physics), Component (thermodynamics), Power management, Pentium, Power optimization, Power budget, Embedded system