CC BY
22
ТЕХНИЧЕСКИЕ НАУКИ
МОДЕЛИРОВАНИЕ И УПРАВЛЕНИЕ СТРАТЕГИЯМИ МИНИМИЗАЦИИ ЭНЕРГОПОТРЕБЛЕНИЯ ЦЕНТРА ОБРАБОТКИ
ДАННЫХ
Бабкин О.В.1, Варламов А.А.2, Горшунов Р.А.3, Дос Е.В.4, Кропачев А.В.5, Зуев Д.О.6 Email: [email protected]
1Бабкин Олег Вячеславович - стратегический консультант, IBM;
2Варламов Александр Александрович - технический директор, ООО "Шаркс Датацентр", г. Москва;
3Горшунов Роман Александрович - архитектор решений, AT&T, г. Братислава, Словакия;
4Дос Евгений Владимирович - ведущий DevOps архитектор, EPAM, г. Минск, Республика Беларусь;
5Кропачев Артемий Васильевич - главный ИТ архитектор, Li9 Technology Solutions, Северная Каролина;
6Зуев Денис Олегович - независимый международный эксперт, Нью Джерси, Соединенные Штаты Америки
Аннотация: проанализированы методики минимизации объемов энергопотребления центров обработки данных путем получения оптимальных пропорций между электропотреблением системы охлаждения и вычислительной системы. Показано, что проблема ограничения энергоэффективности как основного ограничительного фактора для работы центров обработки данных может быть решена путем разработки моделирования динамики термо-профиля серверного зала. Проанализированы гибридные охладительные системы, которые сегодня широко используются в центрах обработки данных. Внедрение гибридных схем требует проведения анализа структуры виртуальных машин и использования более высокого уровня охлаждения благодаря высокой рабочей температуре активных серверов. Разработано решение для оптимизации гибридной архитектуры охлаждения, что позволяет достичь общей минимизации потерь мощности, удовлетворяя базовые требования, указанные в соглашении об уровне обслуживания центра обработки данных. Предложенное решение значительно расширяет удобство охлаждения для центров обработки данных, учитывая климатические условия, нагрузки серверов, температурный режим сервера и архитектуру системы охлаждения сервера. Было продемонстрировано, что для разработки модели оптимизации энергопотребления необходимо оценить оптимальный режим охлаждения и максимальное энергопотребление активных серверов. Для определения графика работы элементов системы охлаждения необходимо оценить энергопотребление центров обработки данных, базовые расходы, количество серверов, виртуальных машин и их размещения. Было проведено сравнение трех типов режима охлаждения для центров обработки данных: фиксированный температурный режим, P-адаптивный режим и PT-адаптивный режим. Фиксированный режим температуры, как обычный режим охлаждения, который использует свободное охлаждение только тогда, когда исходная температура ниже заданной температуры, оказался неэффективным, тогда как P-адаптивный режим и PT-адаптивный режим могут быть использованы в рамках разработанной методологии. Модель оценки эффективности режима охлаждения позволила рекомендовать PT-адаптивную схему, в наибольшей степени оптимизирует процесс потребления энергии и уменьшает затраты на переключение между режимами охлаждения.
Ключевые слова: центр обработки данных, потребление электроэнергии, виртуальная машина, режим свободного охлаждения, режим электро-охлаждения, режим фиксированной температуры, PT адаптивный режим.
MODELING AND CONTROL OF DATA CENTER POWER CONSUMPTION MINIMIZATION STRATEGIES
12 3
Babkin O.V. , Varlamov A.A. , Gorshunov R.A. , Dos E.V.4, Kropachev A.V.5, Zuev D.O.6
'Babkin Oleg Vyacheslavovich - Strategy Consultant, IBM;
2Varlamov Aleksandr Aleksandrovich - Technical Director SHARXDC LLC, MOSCOW;
3Gorshunov Roman Aleksandrovich - Solution Architect, AT&T, BRATISLAVA, SLOVAKIA;
4Dos Evgenii Vladimirovich - Lead DevOps Architect, EPAM, MINSK, REPUBLIC OF BELARUS;
5Kropachev Artemii Vasilyevich - Principal Architect, LI9 TECHNOLOGY SOLUTIONS, NORTH CAROLINA;
6Zuev Denis Olegovich - Independent Consultant, NEW JERSEY, UNITED STATES OF AMERICA
Abstract: оptimization of data center power utilization by getting a proper proportion of computing and cooling power consumption reducing was discussed. It was shown that energy-efficiency constraints problem as a main limiting factor for data centers performance can be solved by thermal modeling and control solutions development. Hybrid cooling solutions which nowadays widely used in a data centers were analyzed. Hybrid cooling schemes require to provide analysis of virtual machines structure organization and utilize higher cooling capability due to the high operating temperature of active servers. Computational and cooling power consumption optimization solution was developed for hybrid cooling architecture. It allows to achieve overall power loss minimization with satisfying of service-level agreement requirements. Proposed solution significantly extends the usability of free cooling for data centers, while it takes into account climate condition, servers' workload, server room's temperature profile and server cooling architecture. It was demonstrated that for development joint power consumption optimization model it is necessary to estimate optimal cooling mode regime and maximum power consumption of active servers. For determination of chillers work schedule it is necessary to estimate power consumption of datacenter, cooling mode transition overheads, number of servers, virtual machines and its placement. There were compared three cooling mode solutions for data centers: fixed temperature regime, P-adaptive regime and PT-adaptive regime. Fixed temperature regime as conventional cooling mode which uses free cooling only when output temperature is lower than pre-defined temperature was proved to be inefficient while P-adaptive regime and PT-adaptive regime was proved to be preferable ones. Developed model of cooling mode efficiency estimation allowed to recommend PT-adaptive regime as adaptive mode which jointly optimizes the power consumption and transition overhead.
Keywords: data center, power consumption, virtual machine, free cooling mode, electrical cooling mode, fixed temperature regime, PT-adaptive regime.
1. Introduction
Data center power utilization level can be optimized by getting a proper proportion of computing and cooling power consumption reducing. Usually conventional computing power minimization solutions lead to actual CPU utilization increase and require higher cooling capability
in order to work with increased heat density of active servers. Nowadays hybrid cooling solutions are widely used in a data centers [1-3], which requires to provide further analysis of virtual machines (VM) structure organization and reduces the chance of using free cooling; hybrid cooling schemes usually require high cooling capability due to the high operating temperature of active servers [4-8].
Therefore joint computational and cooling power consumption optimization solution for data centers was proposed (Figure 1). It was developed for hybrid cooling architecture and allows to achieve overall power loss minimization with satisfying of service-level agreement (SLA) requirements. Proposed solution significantly extends the usability of free cooling for data centers having a hybrid cooling architecture. It takes into account all input parameters of data center server room infrastructure, specifically climate condition, servers' workload, server room's temperature profile and server cooling architecture. Statistics of the climate condition and servers' workload [9-11, 19] forms artificial neural network (ANN) training dataset and further can be forecasted by predictive control scheme. In other hand server room's temperature profile and the dependency between the server temperature and cooling solutions can be simulated and modeled by standard algorithms [12-18].
Fig. 1. Joint power consumption optimization scheme for hybrid cooling architecture
To identify the main aspects of the joint power consumption optimization model development, systematic analysis of recent studies and publications was done. There were analyzed modern cooling solutions for cloud services [1-3], VM placement schemes and free cooling system check [4-8]. Statistics of the climate condition and servers' workload [9-11] allowed forming requirement or predictive control scheme development. Computational algorithms that can be used for server room's temperature profile simulation [12-18] were also discussed.
2. Proposed method
To build joint power consumption optimization model it is necessary to estimate optimal cooling mode regime and maximum power consumption of active servers. It allows to determine
chillers work schedule in order to have no overheads in terms of power and time. Main parameters to be analyzed are:
• power consumption of datacenter PDC;
• cooling mode transition overheads PTR;
• number of servers Ns;
• number of VMs NVM;
• binary matrix representing VM placement B(i, j);
• binary parameter CM which determines cooling mode (CM = 0 for electrical cooling and CM = 1 for free cooling).
Power consumption of datacenter can be calculated as
PDC = PCool^PComp,
(1)
where PCool refers to power consumption of cooling system and PComp represents computational power consumption.
Mathematically the problem solving aspects can be formulated as determining of B(i,j) and CM parameter by obtaining minimums of objective function of entire power consumption PE (Figure 2):
Bii.j) =
¿(1,1)
¿(1,1)
lb(Ns,l) ... b(Ns,NVM). CM £ [0; 1] pi = pdc + ptr
(2)
Fig. 2. Mathematical algorithm ofjoint power consumption scheme for hybrid cooling architecture
It has to be noticed that obtaining minimums of objective function minimum should also include analysis of boundary conditions for server temperature and server performance.
For boundary conditions estimation on should found which refers to temperature of -th server and probability PR Q of actual execution time t A exceeding required execution time tR:
f Ts(i)<Trx m
, (3)
where is maximum temperature constraint for data center servers and is SLA
requirement parameter. Thereby optimization problem can be translated into a bin-packing problem by exploiting the analogy between a bin and a server.
3. Experimental results and analysis
To simplify developed methodology two-phase algorithm can be proposed. It includes determination of optimal pair of parameters of cooling regime and active servers' utilization threshold level { CM; UjH] which allows to satisfy temperature and performance requirements (3). At the second stage VMs have to be assigned to servers in order to minimize number of servers. Optimization procedure should be iterated at every predefined time interval. Thereby equations (2) and (3) could be estimated as:
B{i,j) £ [0, UjH]; CM £ [0; 1] Pz(k) = ZIT'1 ~k ■ (Pcooi + Pcomp + P-"R^ (4)
and boundary conditions could be defined as:
[ U™Q) > Dr/Ns 11 e[k,k + Nk- 1 ] ,
(5)
(U™ < min (Ursnax,UÏ(l)) [ VI, 1< a <0 , ( )
where
- is number of time periods;
- PcRoh Pcomp and PCR are predicted values of Pcooh Pcomp and PTR values at the /-th time period;
- is the prediction of average user requests normalized with maximum number of user requests for single server;
- is a weighting factor;
- is maximum acceptable performance loss regime power;
- is highest utilization satisfying maximum temperature constraint.
Figure 3 shows dependence of the power consumption on the U™ value. Figure demonstrates that total power consumption at free cooling mode is usually changes in proportion to computing power as increase is more significant than the cooling power consumption growth. However, the cooling capability of the free cooling is limited, and maximal value of for this cooling mode is also limited.
Power consumption
Electrical cooling
Computing
Free cooling
Utilization per server
Free cooling
Computing
Fig. 3. Dependence of the power consumption on the active servers' utilization threshold level for free cooling
and electrical cooling regimes
To evaluate the effectiveness of the joint optimization model statistical dataset of CloudSim simulation [19] was used (Figure 4). There were compared three cooling mode solutions for data centers:
• fixed temperature regime as conventional cooling mode which uses free cooling only when output temperature is lower than pre-defined temperature (UjH = U™ax);
• P-adaptive regime as adaptive mode which adjusts the cooling and the utilization threshold to minimize power consumption of data center;
• PT-adaptive regime as adaptive mode which jointly optimizes the power consumption and transition overhead.
The highest power consumption savings were observed at comparison of fixed temperature regime and P-adaptive regime while output temperature was usually higher pre-defined temperature so free cooling mode utilization was usually impossible. In other hand PT-adaptive regime allowed to use free cooling mode by lowering the maximum server power consumption.
Comparison of P-adaptive and PT-adaptive regimes has shown almost similar level of power consumption savings. However, PT-adaptive regime allowed to significantly decrease number of cooling modes transitions by accounting for the overhead caused by the cooling mode transitions. It is important to notice that the effectiveness of PT-adaptive gets enhanced as the power consumption proportion of servers gets to be improved.
Normalized power consumption as the power proportionality of servers should be defined as the ratio of the static power PST to the total power consumption Ps. For low value of PST/PS can be used free cooling for longer periods of time due to lower server utilization threshold. Thereby minimal number of active servers can be used and developed methodology can be used to achieve higher energy-proportionality. Experiments' simulation statistics datasets demonstrates that PT-adaptive allows to provide higher level of power consumption savings for data center infrastructure.
Fig. 4. Comparison of power utilization of fixed temperature (a) and PT-adaptive (b) cooling modes at temperature regime "1" (c) and power utilization ofP -adaptive (d) and PT-adaptive (e) cooling modes at
temperature regime "2 " f
Thereby energy-efficiency constraints problem as a main limiting factor for data centers performance can be solved by thermal modeling and control solutions development which have to be considered as key aspect of power consumption reducing.
4. Conclusions
Key aspects of data center power utilization optimization by getting a proper proportion of computing and cooling power consumption reducing were analyzed. Energy-efficiency constraints problem have to be be solved by thermal modeling and control solutions development. While nowadays hybrid cooling solutions are widely used in a data centers it is important to provide analysis of virtual machines structure organization and utilize higher cooling capability due to the high operating temperature of active servers. It allows to achieve overall power loss minimization with satisfying all requirements.
Proposed methodology extends the usability of free cooling for data centers. It takes into account climate condition, servers' workload, server room's temperature profile and server cooling architecture. It was demonstrated that for development joint power consumption optimization model it is necessary to estimate optimal cooling mode regime and maximum power consumption
of active servers. For determination of chillers work schedule were estimated power consumption of datacenter, cooling mode transition overheads, number of servers, virtual machines and its placement. There were compared three cooling mode solutions for data centers: fixed temperature regime, P-adaptive regime and PT-adaptive regime. Fixed temperature regime was shown as inefficient one while P-adaptive regime and PT-adaptive regime was proved to be preferable ones. Developed model of cooling mode efficiency estimation demonstrated preferences of PT-adaptive regime as adaptive mode which jointly optimizes the power consumption and transition overhead.
References / Список литературы
1. Garday D. "Reducing data center energy consumption with wet side economizers."White paper, Intel, 2007.
2. Atwood D. and Miner J.G. "Reducing data center cost with an air economizer, "White Paper: Intel Corporation, 2008.
3. Lu Т., Lu Х., Remes М. and Viljanen М. "Investigation of air management and energy performance in a data center in Finland: Case study. "Energy and Buildings 43. № 12 (2011): 3360-3372.
4. Kourai K. & Ooba H., 2016. VMBeam: Zero-Copy Migration of Virtual Machines for Virtual IaaS Clouds. 2016 IEEE 35th Symposium on Reliable Distributed Systems (SRDS). D. Kusic, J.O. Kephart, J.E. Hanson, N. Kandasamy and G. Jiang, "Power and performance management of virtualized computing environments via lookahead control, "Cluster computing 12. № 1, 2009: 1-15.
5. Dhiman G., Marchetti G. and Rosing Т. "vGreen: a system for energy efficient computing in virtualized environments," in Proceedings of the 14th ACM/IEEE international symposium on Low power electronics and design. Pp. 243-248. ACM, 2009.
6. Xu J. and Fortes J.A. "Multi-objective virtual machine placement in virtualized data center environments," in Green Computing and Communications (Green Com), 2010 IEEE/ACM Int'l Conference on and Int'l Conference on Cyber, Physical and Social Computing (CPS Com). Pp. 179-188. IEEE, 2010.
7. Jang J.-W., Jeon М., Kim H.-S., Jo Н., Kim J.-S. and Maeng S. "Energy reduction in consolidated servers through memory-aware virtual machine scheduling, "Computers, IEEE Transactions on 60. no. 4 (2011): 552-564.
8. Verma А. et al. "Server workload analysis for power minimization using consolidation," in Proc. USENIX, 2009.
9. Meng Х. et al. "Efficient resource provisioning in compute clouds via VM multiplexign," in Proc. ICAC, 2010.
10. 10. Halder К. et al. "Risk aware provisioning and resoource aggregation based consolidation of virtual machines," in Proc. Cloud, 2012.
11. Weller B., 2010. Installing and Setting Up SQL Server Modeling. Beginning SQL Server Modeling. 1-12.
12. Rivoire S., Ranganathan Р. and Kozyrakis С. "A Comparison of High-Level Full-System Power Models,"HotPower8, 2008: 3-3.
13. Pedram М. and Hwang I. "Power and performance modeling in a virtualized server system," in Parallel Processing Workshops (ICPPW), 2010 39th International Conference on. Pp. 520-526. IEEE, 2010.
14. Patterson М.К. "The effect of data center temperature on energy efficiency," in Thermal and Thermomechanical Phenomena in Electronic Systems, 2008. ITHERM 2008. 11th Intersociety Conference on, pp. 1167-1174. IEEE, 2008.
15. Choi J., Kim Y., Sivasubramanjam А., Srebric J., Wang Q. and Lee J. "A CFD-based tool for studying temperature in rack-mounted servers, "Computers, IEEE Transactions on 57. № 8, 2008: 1129-1142.
16. Brown W. & Lim T., 2017. Quantifying Bolt Relaxation During High Temperature Operation. Volume 3A: Design and Analysis.
17. Nakamura H., 2009. Cooling Fan Model for Thermal Design of Compact Electronic Equipment: Improvement of Modeling Using PQ Curve. ASME 2009 InterPACK Conference, Volume 2.
18. Ayoub R., Nath R. and Rosing T. "JETC: Joint energy thermal and cooling management for memory and CPU subsystems in servers," in High Performance Computer Architecture (HPCA), 2012 IEEE 18th International Symposium on. Pp. 1-12. IEEE, 2012.
19. Buyya R., Ranjan R. and Calheiros R.N. "Modeling and simulation of scalable Cloud computing environments and the CloudSim toolkit: Challenges and opportunities," in High Performance Computing and Simulation, 2009. HPCS'09. International Conference on. Pp. 111. IEEE, 2009.
КРОССПЛАТФОРМЕННЫЙ КОМПИЛЯТОР Караваева О.В.1, Тимин В.С.2 Email: [email protected]
1Караваева Ольга Владимировна - старший преподаватель; 2Тимин Владислав Сергеевич - студент, кафедра электронных вычислительных машин, Вятский государственный университет, г. Киров
Аннотация: стремительное развитие вычислительной техники и, в частности, операционных систем позволяет все более широкому кругу пользователей расширить применение возможностей персонального компьютера. Но для профессиональной разработки системных приложений нужны глубокие знания. Обучение студентов в современных условиях сопряжено с определенными трудностями, когда требуется рассмотреть процессы компиляции, выполняющиеся на самом низком уровне. В рамках данной статьи рассматривается решение данной проблемы с помощью кроссплатформенного компилятора, работающего на виртуальной машине. Ключевые слова: компилятор, грамматика, команда.
CROSS-PLATFORM COMPILER 12 Karavaeva O.V. , Timin V.S.
1Karavaeva Olga Vladimirovna - Senior Teacher;
2Timin Vladislav Sergeevich - Student, DEPARTMENT OF ELECTRONIC COMPUTERS, VYATKA STATE UNIVERSITY, KIROV
Abstract: the rapid development of computer technology and, in particular, of operating systems allows an ever wider range of users to expand the use of the capabilities of a personal computer. But for professional development of system applications you need deep knowledge. Teaching students in modern conditions involves certain difficulties when it is required to consider the compilation processes that are performed at the lowest level. In the framework of this article, we will solve this problem using a cross-platform compiler running on a virtual machine. Keywords: compiler, grammar, command.
УДК 004.431.4
В связи с развитием вычислительной техники появилось множество средств, упрощающих и автоматизирующих разработку программного обеспечения (ПО). Но знание основных принципов работы компьютера всё также важно для разработки системного ПО,
36
