Compile-time dynamic voltage scaling settings: opportunities and limits

With power-related concerns becoming dominant aspects of hardware and software design, significant research effort has been devoted towards system power minimization. Among run-time power-management techniques, dynamic voltage scaling (DVS) has emerged as an important approach, with the ability to provide significant power savings. DVS exploits the ability to control the power consumption by varying a processor's supply voltage (V) and clock frequency (f). DVS controls energy by scheduling different parts of the computation to different (V, f) pairs; the goal is to minimize energy while meeting performance needs. Although processors like the Intel XScale and Transmeta Crusoe allow software DVS control, such control has thus far largely been used at the process/task level under operating system control. This is mainly because the energy and time overhead for switching DVS modes is considered too large and difficult to manage within a single program.In this paper we explore the opportunities and limits of compile-time DVS scheduling. We derive an analytical model for the maximum energy savings that can be obtained using DVS given a few known program and processor parameters. We use this model to determine scenarios where energy consumption benefits from compile-time DVS and those where there is no benefit. The model helps us extrapolate the benefits of compile-time DVS into the future as processor parameters change. We then examine how much of these predicted benefits can actually be achieved through optimal settings of DVS modes. This is done by extending the existing Mixed-integer Linear Program (MILP) formulation for this problem by accurately accounting for DVS energy switching overhead, by providing finer-grained control on settings and by considering multiple data categories in the optimization. Overall, this research provides a comprehensive view of compile-time DVS management, providing both practical techniques for its immediate deployment as well theoretical bounds for use into the future.

[1]  Ulrich Kremer,et al.  The design, implementation, and evaluation of a compiler algorithm for CPU energy reduction , 2003, PLDI '03.

[2]  Mahmut T. Kandemir,et al.  Energy-conscious compilation based on voltage scaling , 2002, LCTES/SCOPES '02.

[3]  Brian W. Kernighan,et al.  AMPL: A Modeling Language for Mathematical Programming , 1993 .

[4]  Krishnendu Chakrabarty,et al.  Investigating the effect of voltage-switching on low-energy task scheduling in hard real-time systems , 2001, ASP-DAC '01.

[5]  Niraj K. Jha,et al.  Power-profile driven variable voltage scaling for heterogeneous distributed real-time embedded systems , 2003, 16th International Conference on VLSI Design, 2003. Proceedings..

[6]  Dirk Grunwald,et al.  Using IPC Variation in Workloads with Externally Specified R ates to Reduce Power Consumption , 2000 .

[7]  Diana Marculescu.  On the Use of Microarchitecture-Driven Dynamic Voltage Scaling , 2000 .

[8]  M. Martonosi,et al.  Run-time power estimation in high performance microprocessors , 2001, ISLPED'01: Proceedings of the 2001 International Symposium on Low Power Electronics and Design (IEEE Cat. No.01TH8581).

[9]  Christopher J. Hughes,et al.  Saving energy with architectural and frequency adaptations for multimedia applications , 2001, MICRO.

[10]  Margaret Martonosi,et al.  Run-time power estimation in high performance microprocessors , 2001, ISLPED '01.

[11]  Ulrich Kremer,et al.  Single Region vs. Multiple Regions: A Comparison of Different Compiler-Directed Dynamic Voltage Scheduling Approaches , 2002, PACS.

[12]  Doug Burger,et al.  Evaluating Future Microprocessors: the SimpleScalar Tool Set , 1996 .

[13]  Gang Qu,et al.  What is the limit of energy saving by dynamic voltage scaling? , 2001, IEEE/ACM International Conference on Computer Aided Design. ICCAD 2001. IEEE/ACM Digest of Technical Papers (Cat. No.01CH37281).

[14]  Miodrag Potkonjak,et al.  MediaBench: a tool for evaluating and synthesizing multimedia and communications systems , 1997, Proceedings of 30th Annual International Symposium on Microarchitecture.

[15]  Margaret Martonosi,et al.  Wattch: a framework for architectural-level power analysis and optimizations , 2000, Proceedings of 27th International Symposium on Computer Architecture (IEEE Cat. No.RS00201).

[16]  Hiroto Yasuura,et al.  Voltage scheduling problem for dynamically variable voltage processors , 1998, Proceedings. 1998 International Symposium on Low Power Electronics and Design (IEEE Cat. No.98TH8379).

[17]  James R. Larus,et al.  Efficient path profiling , 1996, Proceedings of the 29th Annual IEEE/ACM International Symposium on Microarchitecture. MICRO 29.

[18]  Dongkun Shin,et al.  Intra-Task Voltage Scheduling for Low-Energy, Hard Real-Time Applications , 2001, IEEE Des. Test Comput..

[19]  Seongsoo Lee,et al.  Run-time voltage hopping for low-power real-time systems , 2000, DAC.

[20]  A. R. Newton,et al.  Alpha-power law MOSFET model and its applications to CMOS inverter delay and other formulas , 1990 .

[21]  Alan Jay Smith,et al.  Improving dynamic voltage scaling algorithms with PACE , 2001, SIGMETRICS '01.

[22]  Krisztián Flautner,et al.  Automatic Performance Setting for Dynamic Voltage Scaling , 2001, MobiCom '01.

[23]  Thomas D. Burd,et al.  Design issues for Dynamic Voltage Scaling , 2000, ISLPED'00: Proceedings of the 2000 International Symposium on Low Power Electronics and Design (Cat. No.00TH8514).