1
دانشکده مهندسی ، دانشگاه ﻓﺮدوﺳی ﻣﺸﻬﺪ، ﻣﺸﻬﺪ، ایران
2
پژوهشکده علوم کامپیوتر، پژوهشگاه دانش های بنیادین(IPM)، تهران، ایران
چکیده
در سالهای اخیر، دمای بالا و توان مصرفی زیاد در پردازندههای چندهستهای به یک چالش اساسی برای سازندگان و کاربران این پردازندهها تبدیل شده است. با رشد دمای پردازنده، هزینههای خنکسازی و مصرف توان افزایش یافته و طول عمر پردازنده کاهش مییابد. در این تحقیق، یک الگوریتم مدیریت دمای پویا در سطح سیستم عامل پیشنهاد شده است که در اجرای برنامههای موازی مبتنی بر زمانبند ربایش کار، دمای پردازنده را در محدودیت درخواستی کاربر مدیریت میکند. از این رو، ما دو مدل دمایی و کارایی را جهت پیشبینی دمای آینده و تخمین میزان تغییرات کارایی برنامه پیشنهاد دادیم. با استفاده از مدلهای پیشنهادی، الگوریتم پیشنهادی تعداد هستههای فعال و فرکانس پردازنده را به نحوی تعیین میکند که دما از محدودیت تعیین شده پایینتر نگه داشته شده و کمترین آسیب ممکن به کارایی برنامه وارد گردد. آزمایشات بر روی سیستم واقعی نشان داد که الگوریتم پیشنهادی به طور میانگین 28 درصد کارایی بالاتری از الگوریتم آگاه از همسایگی داشته و برخلاف این الگوریتم، هرگز از محدودیت دمایی تعیین شده تخطی نمیکند.
[1] J. Kong, S. W. Chung, and K. Skadron, "Recent thermalmanagement techniques for microprocessors,"In ComputingSurveys (CSUR), vol. 44, no. 3, pp.1-42, 2012.
[2] J. Diaz, C. Munoz-Caro, and A. Nino, "A survey ofparallel programming models and tools in the multi andmany-core era," IEEE Transactions on parallel anddistributed systems, vol. 23, no. 8, pp. 1369-1386, 2012.
[3] S. Zhuravlev, J. C.Saez, Blagodurov, S. Fedorova, andM. Prieto, "Survey of energy-cognizant schedulingtechniques," IEEE Transactionson Parallel and DistributedSystems, vol. 24, no. 7, pp. 1447-1464, 2013.
[4] T. van Dijk, and C. Jaco, "Lace: non-blocking split dequefor work-stealing," In European Conference on ParallelProcessing, pp. 206-217, 2014.
[5] A. Morrison, and Y. Afek, "Fence-free work stealing onbounded TSO processors," ACM SIGPLAN, vol. 49, no. 4,pp. 413-426, 2014.
[6] J. Nakashim, S.Nakatani, and K. Taura, "Design andimplementation of a customizable work stealing scheduler,"In Proceedings of the 3rd International Workshop onRuntime and Operating Systems for Supercomputers, pp.1-9, 2013.
[7] M. Wimme, D. Cederma, J. L. Träff, and P. Tsigas, "Work-stealing with configurable scheduling strategies," InACM SIGPLAN, vol. 48, no. 8, pp. 315-316, 2013.
[8] A. Bhattacharjee, and M.Martonosi, "Thread criticalitypredictors for dynamic performance, power, and resourcemanagement in chip multiprocessors," InACM SIGARCHComputer Architecture, vol. 37, no. 3, pp. 290-301, 2009.
[9] D. Hendler, and N. Shavi, "Non-blocking steal-half workqueues," In Proceedings of the twenty-first annualsymposium on Principles of distributed computing, pp. 280-289, 2002.
[10] U. Acar, A. Chargueraud, and M. Rainey, "Schedulingparallel programs by work stealing with private deques," InACM SIGPLAN, vol. 48, no. 8, pp. 219-228, 2013.
[11] D. Hendler, Y. Lev, M. Moir, and N. Shavit, "Adynamic-sized nonblocking work stealing deque,"Technical report, Sun Microsystems, 2005.
[12] S. Imam, and V. Sarkar, "Load balancing prioritizedtasks via work-stealing," In Euro-Par’15, pp. 222-234, 2015.
[13] G. Liu, M. Fan, G. Quan, and M. Qiu, "On-Linepredictive thermal management under peak temperatureconstraints for practical multi-core platforms,"In Journal ofLow Power Electronics, vol. 8, no. 5, pp. 565-578, 2012.
[14] I. Yeo, C. C. Liu, and E. J. Kim, "Predictive dynamicthermal management for multicore systems," In Proceedingsof the 45th annual Design Automation Conference, pp.734-739, 2008.
[15] R. Cochran, and R. Sherie, "Thermal prediction andadaptive control through workload phase detection," ACMTransactions on Design Automation of Electronic Systems,vol. 18, no. 7, 2013.
[16] J. Yang, X. Zhou, M. Chrobak, Y. Zhang, and L. Jin, "Dynamic thermal management through task scheduling,"IEEE International Symposium on Performance Analysis ofSystems and software (ISPASS’08), pp. 191-201, 2008.
[17] A. Merkel, and F. Bellosa, "Task activity vectors: a newmetric for temperature-aware scheduling," In SIGOPSOperating Systems Review, vol. 42, no. 4, pp.1-12, 2008.
[18] O. Sarood, P. Miller, E.Totoni, and L.V. Kale, "LoadBalancing for High Performance Computing Data Centers,"IEEE Transactions on Computers, vol. 61, no. 12, pp.1752-1764, 2012.
[19] O. Sarood, "Optimizing performance under thermal andpower constraints for HPC data centers,"Doctoraldissertation, University of Illinois at Urbana-Champaign,2014.
[20] T. Lu, P. P. Pande, and B. Shirazi,"A Dynamic,Compiler Guided DVFS Mechanism to Achieve Energy-Efficiency in Multi-coreProcessors," Sustainable Computing:Informatics and Systems, vol. 12, pp. 1-9, 2016.
[21] R. A. Shafik, A. Das, S. Yang, G. Merrett, and B. M.Al-Hashimi, "Adaptive energy minimization of OpenMPparallel applications on many-core systems," In Proceedingsof the 6th Workshop on Parallel Programming and Run-TimeManagement Techniques for Many-core Architectures, pp.19-24, 2015.
[22] R. Cochran, C. Hankendi, and A. Coskun, "Identifyingthe optimal energy-efficient operating points of parallelworkloads," In IEEE/ACM International Conference onComputer-Aided Design, pp. 608-615, 2011.
[23] D. D. Sensi, "Predicting Performance and PowerConsumption of Parallel Applications," InternationalConference on Parallel, Distributed, and Network-BasedProcessing, pp. 200-207, 2016.
[24] J. Li, and J. F. Martinez, "Dynamic power-performanceadaptation of parallel computation on chip multiprocessors,"In The Twelfth International Symposium on High-Performance Computer Architecture, pp. 77-87, 2006.
[25] H. F. Sheikh, I. Ahmad, and D. Fan, "An EvolutionaryTechnique for Performance-Energy-Temperature Optimized Scheduling of Parallel Tasks on Multi-Core Processors,"IEEE Transactions on Parallel and DistributedSystems, vol.27, no. 3, pp. 668-681, 2016.
[26] lm-sensors Linux hardware monitoring [Online].Available: http://www.lm-sensors.org, Jan 2017.
[27] Linux cpufreq governors, Linux Kernel [Online].Available:https://www.kernel.org-/doc/Documentation/cpu-freq/governors.txt. Jan 2017.
[28] Intel Cilk Plus [Online]. Available:http://www.cilkplus.org, Jan 2017.