ABSTRACT
A key challenge in next-generation supercomputing is to effectively schedule limited power resources. Modern processors suffer from increasingly large power variations due to the chip manufacturing process. These variations lead to power inhomogeneity in current systems and manifest into performance inhomogeneity in power constrained environments, drastically limiting supercomputing performance. We present a first-of-its-kind study on manufacturing variability on four production HPC systems spanning four microarchitectures, analyze its impact on HPC applications, and propose a novel variation-aware power budgeting scheme to maximize effective application performance. Our low-cost and scalable budgeting algorithm strives to achieve performance homogeneity under a power constraint by deriving application-specific, module-level power allocations. Experimental results using a 1,920 socket system show up to 5.4X speedup, with an average speedup of 1.8X across all benchmarks when compared to a variation-unaware power allocation scheme.
- NASA Advanced Supercomputing Division, NAS Parallel Benchmark Suite v3.3. 2006. http://www.nas.nasa.gov/Resources/Software/npb.html.Google Scholar
- 2013 Exascale Operating and Runtime Systems. Technical report, Advanced Science Computing Research (ASCR), February 2013. http://science.doe.gov/grants/pdf/LAB13-02.pdf.Google Scholar
- AMD. AMD Turbo CORE Technology. http://www.amd.com/us/products/desktop/processors/phenom-ii/Pages/phenom-ii-key-architectural-features.aspx.Google Scholar
- S. Ashby, P. Beckman, J. Chen, P. Colella, B. Collins, D. Crawford, J. Dongarra, D. Kothe, R. Lusk, P. Messina, T. Mezzacappa, P. Moin, M. Norman, R. Rosner, V. Sarkar, A. Siegel, F. Streitz, A. White, and M. Wright. The Opportunities and Challenges of Exascale Computing. 2010.Google Scholar
- P. Bailey, D. Lowenthal, V. Ravi, B. Rountree, M. Schulz, and B. de Supinski. Adaptive Configuration Selection for Power-Constrained Heterogeneous Systems. In International Conference on Parallel Processing, ICPP '14, 2014.Google Scholar
- D. Bodas, J. Song, M. Rajappa, and A. Hoffman. Simple Power-aware Scheduler to Limit Power Consumption by HPC System Within a Budget. In Proceedings of the 2nd International Workshop on Energy Efficient Supercomputing, pages 21--30. IEEE Press, 2014. Google ScholarDigital Library
- S. Borkar. Designing Reliable Systems from Unreliable Components: The Challenges of Transistor Variability and Degradation. Micro, IEEE, 25(6):10--16, Nov 2005. Google ScholarDigital Library
- S. Borkar, T. Karnik, S. Narendra, J. Tschanz, A. Keshavarzi, and V. De. Parameter Variations and Impact on Circuits and Microarchitecture. In Proceedings of the 40th annual Design Automation Conference, pages 338--342, June 2003. Google ScholarDigital Library
- K. W. Cameron, X. Feng, and R. Ge. Performance-constrained, Distributed DVS Scheduling for Scientific Applications on Power-aware Clusters. In Supercomputing, November 2005. Google ScholarDigital Library
- N. P. Carter, A. Agrawal, S. Borkar, R. Cledat, H. David, D. Dunning, J. B. Fryman, I. Ganev, R. A. Golliver, R. C. Knauerhase, R. Lethin, B. Meister, A. K. Mishra, W. R. Pinfold, J. Teller, J. Torrellas, N. Vasilache, G. Venkatesh, and J. Xu. Runnemede: An Architecture for Ubiquitous High-Performance Computing. In 19th IEEE International Symposium on High Performance Computer Architecture, HPCA 2013, Shenzhen, China, February 23--27, 2013, pages 198--209, 2013. Google ScholarDigital Library
- M. Curtis-Maury, F. Blagojevic, C. D. Antonopoulos, and D. S. Nikolopoulos. Prediction-Based Power-Performance Adaptation of Multithreaded Scientific Codes. IEEE Trans. Parallel Distrib. Syst., 19(10):1396--1410, Oct. 2008. Google ScholarDigital Library
- H. David, E. Gorbatov, U. Hanebutte, R. Khanna, and C. Le. RAPL: Memory Power Estimation and Capping. In Proceedings of the 16th ACM/IEEE international symposium on Low power electronics and design, ISLPED '10, pages 189--194, 2010. Google ScholarDigital Library
- D. DeBonis, J. H. Laros III, and K. Pedretti. Qualification for PowerInsight Accuracy of Power Measurements.Google Scholar
- S. Dighe, S. Vangal, P. Aseron, S. Kumar, T. Jacob, K. Bowman, J. Howard, J. Tschanz, V. Erraguntla, N. Borkar, V. De, and S. Borkar. Within-Die Variation-Aware Dynamic-Voltage-Frequency-Scaling With Optimal Core Allocation and Thread Hopping for the 80-Core TeraFLOPS Processor. Solid-State Circuits, IEEE Journal of, 46(1):184--193, Jan 2011.Google Scholar
- D. A. Ellsworth, A. D. Malony, B. Rountree, and M. Schulz. POW: System-wide Dynamic Reallocation of Limited Power in HPC. June 2015.Google ScholarDigital Library
- H. Esmaeilzadeh, E. Blem, R. S. Amant, K. Sankaralingam, and D. Burger. Power Challenges May End the Multicore Era. Commun. ACM, 56(2):93--102, Feb. 2013. Google ScholarDigital Library
- H. Esmaeilzadeh, E. Blem, R. St. Amant, K. Sankaralingam, and D. Burger. Dark Silicon and the End of Multicore Scaling. In Proceedings of the 38th Annual International Symposium on Computer Architecture, ISCA '11, pages 365--376, New York, NY, USA, 2011. ACM. Google ScholarDigital Library
- M. Etinski, J. Corbalan, J. Labarta, and M. Valero. Optimizing Job Performance Under a Given Power Constraint in HPC Centers. In Green Computing Conference, pages 257--267, 2010. Google ScholarDigital Library
- M. Etinski, J. Corbalan, J. Labarta, and M. Valero. Linear Programming Based Parallel Job Scheduling for Power Constrained Systems. In International Conference on High Performance Computing and Simulation, pages 72--80, 2011.Google Scholar
- D. Feitelson, U. Schwiegelshohn, and L. Rudolph. Parallel Job Scheduling - A Status Report. In In Lecture Notes in Computer Science, pages 1--16. Springer-Verlag, 2004. Google ScholarDigital Library
- R. Ge, X. Feng, W. Feng, and K. W. Cameron. CPU MISER: A Performance-Directed, Run-Time System for Power-Aware Clusters. In International Conference on Parallel Processing, 2007. Google ScholarDigital Library
- L. R. Harriott. Limits of lithography. Proceedings of the IEEE, 89(3):366--374, 2001.Google ScholarCross Ref
- C. Hsu and W. Feng. A Power-Aware Run-Time System for High-Performance Computing. In Supercomputing, November 2005. Google ScholarDigital Library
- J. Hu, C. J. Xue, Q. Zhuge, W.-C. Tseng, and E.-M. Sha. Towards Energy Efficient Hybrid On-chip Scratch Pad Memory with Non-volatile Memory. In Design, Automation & Test in Europe Conference & Exhibition (DATE), 2011, pages 1--6. IEEE, 2011.Google Scholar
- J. H. L. III, P. Pokorny, and D. Debonis. PowerInsight - A commodity power measurement capability. In IGCC'13, pages 1--6, 2013.Google Scholar
- Intel. Intel Turbo Boost Technology 2.0. http://www.intel.com/content/www/us/en/architecture-and-technology/turbo-boost/turbo-boost-technology.html.Google Scholar
- Intel. Intel-64 and IA-32 Architectures Software Developer's Manual, Volumes 3A and 3B: System Programming Guide. 2011.Google Scholar
- S. Jilla. Minimizing The Effects Of Manufacturing Variation During Physcial Layout. http://chipdesignmag.com/display.php?articleId=2437.Google Scholar
- L. Kalé and S. Krishnan. CHARM++: A Portable Concurrent Object Oriented System Based on C++. In Proceedings of OOPSLA'93. ACM Press, September 1993. Google ScholarDigital Library
- N. Kappiah, V. W. Freeh, and D. K. Lowenthal. Just In Time Dynamic Voltage Scaling: Exploiting Inter-Node Slack to Save Energy in MPI Programs. Journal of Parallel and Distributed Computing, 68:1175--1185, 2008. Google ScholarDigital Library
- T. Karnik, M. Pant, and S. Borkar. Power management and delivery for high-performance microprocessors. In The 50th Annual Design Automation Conference 2013, DAC '13, Austin, TX, USA, May 29 - June 07, 2013, page 159, 2013. Google ScholarDigital Library
- H. Kaul, M. Anders, S. Hsu, A. Agarwal, R. Krishnamurthy, and S. Borkar. Near-threshold Voltage (NTV) Design: Opportunities and Challenges. In The 49th Annual Design Automation Conference 2012, DAC '12, June 2012. Google ScholarDigital Library
- Z. Liu, J. Lofstead, T. Wang, and W. Yu. A case of system-wide power management for scientific applications. In Cluster Computing (CLUSTER), 2013 IEEE International Conference on, September 2013.Google ScholarCross Ref
- P. Luszczek, D. Bailey, J. Dongarra, J. Kepner, R. Lucas, R. Rabenseifner, and D. Takahash. HPC Challenge Benchmark Suite. ttp://icl.cs.utk.edu/pcc/index.html.Google Scholar
- A. Marathe, P. Bailey, D. K. Lowenthal, B. Rountree, M. Schulz, and B. R. de Supinski. A Run-time System for Power-constrained HPC Applications. In In International Supercomputing Conference (ISC, July 2015.Google Scholar
- N. Maruyama, S. Suzuki, K. Mikami, Y. Komuro, S. Takizawa, and M. Matsuda. Fiber Miniapp Suite. fiber-miniapp.github.io.Google Scholar
- S. Miwa, S. Aita, and H. Nakamura. Performance Estimation of High Performance Computing Systems with Energy Efficient Ethernet Technology. Computer Science - Research and Development, 29:161--169, August 2014. Google ScholarDigital Library
- T. Ogino, R. J. Walker, and M. Ashour-Abdalla. A Global Magnetohydrodynamic Simulation of the Magnetopause when the Interplanetary Magnetic Field is Northward. IEEE Transaction on Plasma Science, 20:817--828, December 1992.Google ScholarCross Ref
- T. Patki, D. K. Lowenthal, B. Rountree, M. Schulz, and B. R. de Supinski. Exploring Hardware Overprovisioning in Power-constrained, High Performance Computing. In International Conference on Supercomputing, pages 173--182, 2013. Google ScholarDigital Library
- T. Patki, A. Sasidharan, M. Maiterth, D. Lowenthal, B. Rountree, M. Schulz, and B. de Supinski. Practical Resource Management in Power-Constrained, High Performance Computing. In High Performance Parallel and Distributed Computing (HPDC), June 2015. Google ScholarDigital Library
- B. Rountree, D. H. Ahn, B. R. de Supinski, D. K. Lowenthal, and M. Schulz. Beyond DVFS: A First Look at Performance under a Hardware-Enforced Power Bound. In IPDPS Workshops (HPPAC), pages 947--953. IEEE Computer Society, 2012. Google ScholarDigital Library
- B. Rountree, D. Lowenthal, B. de Supinski, M. Schulz, V. Freeh, and T. Bletch. Adagio: Making DVS Practical for Complex HPC Applications. In International Conference on Supercomputing, June 2009. Google ScholarDigital Library
- B. Rountree, D. Lowenthal, M. Schulz, and B. de Supinski. Practical Performance Prediction Under Dynamic Voltage Frequency Scaling. In International Green Computing Conference, July 2011. Google ScholarDigital Library
- S. Samaan. The Impact of Device Parameter Variations on the Frequency and Performance of VLSI Chips. In Computer Aided Design, 2004. ICCAD-2004. IEEE/ACM International Conference on, pages 343--346, Nov 2004. Google ScholarDigital Library
- V. Sarkar, W. Harrod, and A. Snavely. Software Challenges in Extreme Scale Systems. In Journal of Physics, Conference Series 012045, 2009.Google ScholarCross Ref
- O. Sarood. Optimizing Performance Under Thermal and Power Constraints for HPC Data Centers. PhD thesis, University of Illinois, Urbana-Champaign, December 2013.Google Scholar
- O. Sarood, A. Langer, A. Gupta, and L. V. Kale. Maximizing Throughput of Overprovisioned HPC Data Centers Under a Strict Power Budget. In Supercomputing, Nov. 2014. Google ScholarDigital Library
- S. Shende and A. D. Malony. The Tau Parallel Performance System. IJHPCA, 20(2):287--311, 2006. Google ScholarDigital Library
- K. Shoga, B. Rountree, and M. Schulz. Whitelisting MSRs with msr-safe. November 2014.Google Scholar
- R. Teodorescu and J. Torrellas. Variation-Aware Application Scheduling and Power Management for Chip Multiprocessors. In Computer Architecture, 2008. ISCA '08. 35th International Symposium on, pages 363--374, June 2008. Google ScholarDigital Library
- E. Totoni, A. Langer, J. Torrellas, and L. Kale. Scheduling for HPC Systems with Process Variation Heterogeneity. January 2015.Google Scholar
- J. Tschanz, J. Kao, S. Narendra, R. Nair, D. Antoniadis, A. Chandrakasan, and V. De. Adaptive Body Bias for Reducing Impacts of Die-to-die and Within-die Parameter Variations on Microprocessor Frequency and Leakage. Solid-State Circuits, IEEE Journal of, 37(11):1396--1402, Nov 2002.Google Scholar
- S. Wallace, V. Vishwanath, S. Coghlan, Z. Lan, and M. E. Papka. Measuring power consumption on IBM Blue Gene/Q. In Parallel and Distributed Processing Symposium Workshops & PhD Forum (IPDPSW), 2013 IEEE 27th International, pages 853--859. IEEE, 2013. Google ScholarDigital Library
- S. Wallace, V. Vishwanath, S. Coghlan, J. Tramm, Z. Lan, and M. Papkay. Application power profiling on IBM Blue Gene/Q. In Cluster Computing (CLUSTER), 2013 IEEE International Conference on, pages 1--8. IEEE, 2013.Google ScholarCross Ref
- K. Yoshii, K. Iskra, R. Gupta, P. Beckman, V. Vishwanath, C. Yu, and S. Coghlan. Evaluating power-monitoring capabilities on IBM Blue Gene/P and Blue Gene/Q. In Cluster Computing (CLUSTER), 2012 IEEE International Conference on, pages 36--44. IEEE, 2012. Google ScholarDigital Library
Index Terms
- Analyzing and mitigating the impact of manufacturing variability in power-constrained supercomputing
Recommendations
Practical Resource Management in Power-Constrained, High Performance Computing
HPDC '15: Proceedings of the 24th International Symposium on High-Performance Parallel and Distributed ComputingPower management is one of the key research challenges on the path to exascale. Supercomputers today are designed to be worst-case power provisioned, leading to two main problems --- limited application performance and under-utilization of procured ...
Runtime-Guided Mitigation of Manufacturing Variability in Power-Constrained Multi-Socket NUMA Nodes
ICS '16: Proceedings of the 2016 International Conference on SupercomputingCurrent large scale systems show increasing power demands, to the point that it has become a huge strain on facilities and budgets. Researchers in academia, labs and industry are focusing on dealing with this "power wall", striving to find a balance ...
Power Tuning HPC Jobs on Power-Constrained Systems
PACT '16: Proceedings of the 2016 International Conference on Parallel Architectures and CompilationAs we approach the exascale era, power has become a primary bottleneck. The US Department of Energy has set a power constraint of 20MW on each exascale machine. To be able achieve one exaflop under this constraint, it is necessary that we use power ...
Comments