Welcome to MIST SPAN Site.
SPAN logo
Motivation:
As an indispensable component of a computer system, software has a profound impact on power dissipation. However, information scarcity of dynamic power dissipation impedes the progress of energy-efficient software design. For example, improper usage of system resources in a program is usually difficult to detect. We propose a set of software function level power analysis tools, SPAN. We explore the method to enable instrumentation for power analysis while minimizing overhead. The tool is supposed to provide function level (subroutine call) power profiling once the power model is constructed. As a result, code “hot region” can be spotted for further power optimization without hardware instrumentation. Because the diversity of applications in real world, we use different profiling techniques in order to generate accurate and fine-grained power profile for an application.

Method:
We developed a light-weighted library, which is the original SPAN function APIs to profile a given application. SPAN uses IPC as model input and generate power profile for the insrumented functions. For example, in order to instrument the benchmark of IS in NPB benchmark suite, first, SPAN_initialize() is invoked. Then, use SPAN_start(“function_name”) right before the function entrance. Call SPAN_stop(“function_name”) after the function. Do not forget to close everything at the end, SPAN_stop();
Method

Publications:
Shinan Wang, Hui Chen and Weisong Shi, SPAN: A Software Power Analyzer for Multicore Computer Systems, Elsevier Sustainable Computing: Informatics and Systems, Vol. 1, No.1, pp. 23-34, January 2011.

Hui Chen, Shinan Wang and Weisong Shi, Where Does the Power Go in a Computer System: Experimental Analysis and Implications, accepted by PMP, in conjunction with International Green Computing Conference (IGCC), July 2011.