CC BY
60
НАУКОЕМКИЕ ТЕХНОЛОГИИ В КОСМИЧЕСКИХ ИССЛЕДОВАНИЯХ ЗЕМЛИ, Т 10 № 6-2018
ПУБЛИКАЦИИ НА АНГЛИЙСКОМ ЯЗЫКЕ: ИНФОРМАТИКА, ВЫЧИСЛИТЕЛЬНАЯ ТЕХНИКА И УПРАВЛЕНИЕ
doi: 10.24411/2409-5419-2018-10192
MODELS AND METHODS OF RESOURCE ALLOCATION OF INFOCOMMUNICATION SYSTEM IN CLOUD DATA CENTERS
ANDREW V. TOUTOV
ABSTRACT
The main problem of modern data centers is the colossal power consumption. In order to minimize power consumption it is necessary to improve methods for resources allocation, while ensuring a high quality of service. Reallocation of resources in the cloud data center occurs through live migration of virtual machines, which additionally loads the system and interrupts monitoring of servers. Currently, there is a large number of works devoted to individual issues of optimal allocation and resource management of cloud data centers. However, in known works there is no complete cycle of work. This paper proposes models and methods for a complete cycle of optimization and resource management of the cloud data center infocommunication system. In particular, the model for the initial placement of virtual machines in the form of a multicriteria optimization problem and the method for solving it are proposed. A two-level resource management system is presented, which includes local and global controllers. The local controller monitors the load and temperature of servers and makes a prediction for the next monitoring window. For prediction, it is suggested to use the method of group method of data handling (GMDH). To determine the size of the monitoring window two types of live migration have been studied and the method for calculating total migration time has been proposed, based on finding an analytical expression for the probability density. For SaaS and PaaS services which use horizontal scaling, a model of two-criteria optimization of the number of virtual machines in clusters of a large multi-tier application is proposed. It is suggested to solve this by a combined method of successive concessions and restrictions.
Information about author:
Senior Lecturer of the Moscow technical university of communications and informatics, Moscow, Russia, andrew_vidnoe@mail.ru
KEYWORDS: resource management; cloud computing; virtual machine placement; live migration; data center; resource allocation.
For citation: Tutov A.V. Models and methods of resources allocation of infocommunication system in cloud data centers. H&ES Research. 2018. Vol. 10. No. 6. Pp. 100-00. doi: 10.24411/2409-5419-2018-10192
Vol 10 No 6-2018, H&ES RESEARCH = PUBLICATIONS IN ENGLISH: INFORMATICS, COMPUTER ENGINEERING AND CONTROL
Introduction
In recent years, the activities of telecom operators have undergone a digital transformation, in which telecommunications operators have to provide digital services in data centers. Cloud computing is one of the most important areas. It is based on virtualization technology that allows to implement the basic requirements of cloud services, such as the use of resources on demand and resource elasticity, through the horizontal and vertical scaling of applications and live migration of virtual machines (VM).
The infrastructure and architectural solutions of data centers focused on cloud services are not fundamentally different from traditional virtualized data centers. However, evolution to cloud data centers at the same time puts more stringent requirements for the performance of the infocommunication system, cooling systems and power supplies in a high-density environment.
The main problem of modern data centers is the consumption of colossal amounts of power, a significant part of which goes to auxiliary systems, in particular a cooling system, which begins to work hard in the event of uneven temperature distribution in the hall. In addition, despite the fact that many data centers are already filled, a significant part of the computing resources is used inefficiently.
To minimize power consumption, heat generation, uniform resource utilization and provide service quality according to SLA agreements, it is necessary to improve the resource allocation process, including methods for initial and dynamic VM placement, as well as scaling.
Reallocation of resources in the cloud data center occurs by VM migration, which is a costly operation, additionally loading network and servers, affecting the quality of services and server monitoring. Therefore, in order to perform effective monitoring and forecasting the state of servers, it is necessary to estimate the total migration time of VMs.
Currently, there is a large number of works devoted to individual issues of optimal allocation and management of resources of cloud data centers. The problems of the initial placement are considered in papers [1-6], dynamic resource allocation in [7-11], scaling issues in [12, 13], live migration of virtual machines in [14-16]. However, in known works there is no complete cycle of work on optimization and management of cloud data center resources.
In this paper, proposed models and methods the proposed models and methods constitute a full cycle optimization and resource management of the cloud data center info-communication system, including initial and dynamic virtual machine placement, method for forecasting server overloads and underloads, and method for calculating the duration of VM live migration to select the size of the monitoring window.
Initial placement
Depending on the initial state of the data center, the VM allocation methods are divided into two types: initial and dynamic placement.
The initial placement occurs when a group of virtual machines are located in the data center for the first time or transferred from one data center to another. In most works devoted to optimization of the initial placement, the following statement is used: there are M different physical servers connected to the network, each of which is characterized by the performance of the processor (CPU) and the amount of memory (RAM). Also, the network bandwidth and the load coming to each server are known. In addition, N virtual machines are specified, the characteristics of which are ordered by users. It is required to consolidate virtual machines on physical servers, so that energy consumption, unused resources, uneven heat dissipation and violation of SLA-agreements are minimized. To solve this problem, heuristic algorithms FFD and BFD are proposed, as well as their modifications [1]. Evolution algorithms are also used to obtain results on this problem [3-5].
Multicriteria problem is solved by constructing the generalized criterion [5]. In [6] it was shown that the best result can be achieved using a combined method of successive concessions and restrictions [17].
Horizontal scaling
One of the most important advantages of cloud computing is resource elasticity, i.e. the ability to add resources to the application and pay only for consumed resources. To scale resources in cloud computing, virtual machines that run the same application are clustered. Horizontal scaling is the process of adding or removing VMs from a cluster. Cloud providers of SaaS and PaaS services negotiate into service level agreements (SLA) with users, in which one of the indicators is the average response time for a query or the maximum response time for a given percentage of requests. Currently, the predominant architecture of Web applications is a multi-tier architecture, implying the use of Web servers for data representation, application servers for implementing application logic and database servers for managing databases. Schematically, a cloud with multi-tiered applications is shown in fig. 1.
Such a cloud is modeled as an open queuing network, where each node is a cluster of virtual machines such as m/G/1/ PS. The discipline of service is the processor sharing (PS). For such a network, the expression to determine mean response time was obtained in [12]. Since the number of directly running virtual machines in the cluster can vary depending on the load, it is possible to determine the optimal number of virtual machines in clusters of a multi-tier application by the criteria of minimum costs and maximum capacity:
НАУКОЕМКИЕ ТЕХНОЛОГИИ В КОСМИЧЕСКИХ ИССЛЕДОВАНИЯХ ЗЕМЛИ, Т 10 № 6-2018
ПУБЛИКАЦИИ НА АНГЛИЙСКОМ ЯЗЫКЕ: ИНФОРМАТИКА, ВЫЧИСЛИТЕЛЬНАЯ ТЕХНИКА И УПРАВЛЕНИЕ
Fig. 1. Cloud data center with multi-tier applications.
min ftco(Sl,...,sc) = ZTCOc • Sc;
{S!,...,SC } c=i
max FTP(S1,...,Sc) =
{S1,...,SC} 1
S
к ' K ^ m
K k=i
under constraints
¿ m
c=i Xk sc-qc
< T
SLA'
k = 1,. . ., K, q < S < S "
" 7 7 ±c с — с
с = 1,
С.
where C — the number of virtual clusters of data centers; K —the number of query classes; S — number of servers in the cluster c; FTp—criterion of the data center throughput; FTCO—criterion of total cost of ownership of data center servers;
TCOc — total cost of ownership of one cluster server c; q — nominal cluster load with one server at load 1; mkc — the average processing time of a class k query by a
weakly loaded cluster server c;
y k
Ac — the average number of visits to the cluster c with r the
query of class k during its time in the system;
Xc — the arrival rate of query of class k in the cluster c;
1—the arrival rate of query of class k in the system;
Tsla — restriction on the average response time for a query of class k specified in the SLA-agreement;
Sc"° — the maximum possible number of servers in the cluster c.
This problem is proposed to be solved by a combined method of successive concessions and restrictions [17].
Dynamic allocation
Data centers must provide sufficient resources for the hosted applications. The operating conditions can be characterized by significant load changes. In cases where these changes affect application performance, the resource management system has to dynamically reallocate resources by redistributing virtual machines between physical servers without loss of VM availability. For cloud providers, the most important criteria are server power consumption, cooling costs, and uniform resource utilization.
Dynamic resource allocation is implemented by the resource management system and is based on the use of virtual machine migration depending on the current operating conditions of the system. The resource management system should answer the following questions:
• When to migrate VMs?
• Which VMs to choose for migration?
• Where to migrate the VMs?
Vol. 10. No. 6-2018, H&ES RESEARCH PUBLICATIONS IN ENGLISH: INFORMATICS, COMPUTER ENGINEERING AND CONTROL
The cloud resource management system has a two-tier architecture consisting of global and local controllers (fig. 2) [7].
Local controllers analyze the state of the physical servers on which they are located and determine the possible underload, overload and overheat states based on the forecast for the next observation window, which is performed using the group method of data handling (GMDH). If one of the following conditions is detected, the local controller reports this to the global controller, which selects the destination server to which the VM will be migrated.
In this system, local controllers decide which VM and when it should be moved. While global controller answers the question where to move the VM. To do this, the local controllers constantly analyze the monitoring system data, and in case of output of the working model parameters beyond the permissible level, they inform the global controller that activates the migration process of virtual machines.
Verification of the system indicators is carried out sequentially, in accordance with the importance of the criteria (overheating, overload, underload of the host). The monitoring process is carried out continuously, even during the migration of the virtual machine. Priority criteria can be changed by the system administrator, but due to their inconsistency, the verification phase should remain sequential (fig. 3).
In case of condition for VM migration, the global controller determines the VM on the problem hosts and starts the algorithm for finding the optimal destination host. The destination host selection is also a sequential step, where each application is processed in the queue order. After determining the destination host and starting the migration process, this physical server is temporarily removed from the list of available nodes until the end of migration.
The process of dynamic placement of virtual machines includes monitoring and forecasting the state of servers. As a method of forecasting, it is suggested to use the group method of data handling (GMDH) [7, 18].
It was noted that the criterion for choosing the optimal structure of the model strongly affects the quality of forecasting. The best results were obtained with the choice of the model by minimizing the criterion of regularity at the last two points of the examination sample. This allows us to take into account the latest sample data for a more accurate short-term forecast. To test the effectiveness of the dynamic allocation of resources with the proposed forecasting method, a simulation model was developed using the CloudSim package [19]. It is shown that the GMDH combinatorial algorithm for predicting the characteristics of destination hosts for all samples yields
Fig. 2. Two-tier system for managing the resources of the cloud data center
НАУКОЕМКИЕ ТЕХНОЛОГИИ В КОСМИЧЕСКИХ ИССЛЕДОВАНИЯХ ЗЕМЛИ, Т. 10. № 6-2018
ПУБЛИКАЦИИ НА АНГЛИЙСКОМ ЯЗЫКЕ: ИНФОРМАТИКА, ВЫЧИСЛИТЕЛЬНАЯ ТЕХНИКА И УПРАВЛЕНИЕ
Fig. 3. The algorithm of controller operation
results not worse than those published earlier, obtained by the local regression method [8].
The quality of forecast depends on the correctness of the data collected by the monitoring subsystem. Since migration further burdens the network and servers on both sides, it is necessary to estimate the migration time, which has a wide spread depending on the application [20]. In [21] it was shown that in order for migration not to affect the monitoring process for servers, the size of the monitoring window should be larger than the total migration time of the migration. A method is proposed for calculating the total migration time of migration on the basis of finding the analytical expression for the probability density of total migration time.
Method for calculation the total migration time
The calculation method is justified in [22] and consists of a number of stages.
1. Collect data on the number of migrations N and the number of elementary operations in the migration X for the periods of observations T. The elementary operation of the migration process is the minimum deterministic part of it, denoted t .
TR — time for resource reservation;
TSC — time for stop and copy phase;
TC — time for commitment;
TA — time for activation.
2. Estimate the parameters of the distributions of the number of migrations and the number of elementary operations
X — migration arrival rate (the parameter of Poisson distribution);
p, o2 — mathematical expectation the variance of the relative number of elementary operations (parameters of normal distribution);
For post-copy migration lgX should be taken instead of X.
3. Construct the time dependences of the parameters a(T), o(T) and X(T).
4. Using the method of least squares, find the values of the coefficients y and P in the models.
a(T) = Yi • T(1
_ e-Pix)
k(T) = Y2 • T(1 - x )
^min = TP + TR + TSC + TC + TA ,
where Т„ — the initialization time;
a2(T) = y3 • T(1 - е~взx )
Vol 10 No 6-2018, H&ES RESEARCH = PUBLICATIONS IN ENGLISH: INFORMATICS, COMPUTER ENGINEERING AND CONTROL
5. Find the ratios of the coefficients obtained
c =
II.
Y 2 '
c2 =
II. Y2
6. Calculate b by formula
2^/c
b =
7. Substitute the obtained coefficients in the expression for the distribution density of the migration total migration time.
The general algorithm of the method is shown in fig. 4. To simplify the calculations, it is proposed to use the Charlier A-series, which is acceptable for the studied volume of statistical data.
The obtained expression allows us to determine with certain probability the window closure criterion on the local controllers in the data center resource management system, which is important for effective system monitoring.
Conclusion
In this work the full cycle of works on the resource management of the infocommunication system of cloud data centers is proposed, including models and methods of initial and dynamic placement of virtual machines and horizontal scaling. It takes into account the VM live migration, which, in comparison with other models and methods, allows maintaining the stability and quality of cloud services for infocommu-nication system.
For forecasting it is proposed to use the group method of data handling, which allows to increase the accuracy of the server overload forecast, to reduce the number of unnecessary migrations and to increase the stability of the cloud data center as a whole.
A method for calculating the total migration time is proposed, based on finding an approximate analytical expression for the probability density, which makes it possible to determine with certain probability the window closure criterion on local controllers in the data center resource management system.
References
1. Verma A., Ahuja P., Neogi A. pMapper: power and migration cost aware application placement in virtualized systems. Proceedings of the 9th ACM/IFIP/USENIX International Conference on Middleware (Leuven, Belgium, December 1-5, 2008). New York, 2008. Vol. 5346. Pp. 243-264.
2. Mikryukov A. A., Hantimirov R. Initial resource pro-visioningin iaas clouds based on the analytic hierarchy process. Statistics and Economics. 2015. No. 4. Pp.184-187. URL: https://doi.org/10.21686/2500-3925-2015-4-184-187. (In Russian)
Fig. 4. The algorithm for obtaining the probability density of total migration time
3. Kostenko V. A., Plakunov A. V. Ant Algorithms for Scheduling Computations in Data Processing Centers. Moscow University Computational Mathematics and Cybernetics. 2017. Vol. 41. No.1. Pp. 44-50.
4. Alharbi F., Tian Y. C., Tang M., Ferdaus M. H. Profile-Based Ant Colony Optimization for Energy-Efficient Virtual Machine Placement. Proc. of the International Conference on
c
2
НАУКОЕМКИЕ ТЕХНОЛОГИИ В КОСМИЧЕСКИХ ИССЛЕДОВАНИЯХ ЗЕМЛИ, Т 10 № 6-2018
ПУБЛИКАЦИИ НА АНГЛИЙСКОМ ЯЗЫКЕ: ИНФОРМАТИКА, ВЫЧИСЛИТЕЛЬНАЯ ТЕХНИКА И УПРАВЛЕНИЕ
Neural Information Processing, ICONIP 2017(Guangzhou, China, November 14-18, 2017). In: Liu D., Xie S., Li Y., Zhao D., El-Alfy E.S.M. (eds). Springer, 2017. Vol. 10634. Pp. 863-871.
5. Xu J., Fortes J. Multi-objective Virtual Machine Placement in Virtualized Data Center Environments. Proceedings of the Green Computing and Communications (GreenCom), 2010 IEEE/ACMInt'l Conference on & Int'l Conference on Cyber, Physical and Social Computing (CPSCom), Hangzhou, 18-20 December 2010. IEEE, 2010. Pp. 179-188.
6. Vorozhtsov A. S., Tutova N. V., Tutov A. V. Optimal cloud servers placement in data centers. T-Comm. 2015. Vol. 9. No. 6. Pp. 4-8. (In Russian)
7. Vorozhtsov A. S., Tutova N. V., Tutov A. V. Dynamic computing resource allocation in data centers. T-Comm. 2016. Vol. 10. No.7. Pp. 47-51. (in Russian)
8. Beloglazov A., Buyya R. Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in Cloud data centers. Concurrency and Computation: Practice and Experience (CCPE). 2012. Vol. 24. No. 13. Pp. 1397-1420.
9. Larin A. A., Abrosimov L. I. A Methodology for Redistributing the Operating Virtual Machines among the Servers in a Data Center. VestnikMEI. No. 1. Pp. 98-105. DOI: 10.24160/19936982-2018-1-98-105. (In Russian)
10. Li X., Garraghan P., Jiang X. et al. Holistic Virtual Machine Scheduling in Cloud Datacenters towards Minimizing Total Energy. IEEE Transactions on Parallel and Distributed Systems. 2018. No. 29 (6). Pp. 1317-1331.
11. Radhakrishnan A., Kavitha V. Energy conservation in cloud data centers by minimizing virtual machines migration through artificial neural network. Computing. 2016. Vol. 98. Issue 11. Pp 1185-1202.
12. Vorojcov A. S., Tutova N. V., Tutov A. V. Performance evaluation of cloud data centers. T-Comm. 2014. Vol. 8. No. 5. Pp. 69-71. (In Russian)
13. Kaneko Y., Ito T., Ito M., Kawazoe H. Virtual Machine Scaling Method Considering Performance Fluctuation of Public Cloud. 2017 IEEE10th International Conference on Cloud Computing (CL0UD2014), Honolulu, 25-30june 2017). NY, 2017. Pp. 782-785. doi: 10.1109/CL0UD.2017.114
14. Tikhomirov P. O., Emelyanov P. V., Plotnik N. S., Zyrya-nov A. V. Minimizing downtime processes during their migration in the cloud hosting. Novosibirsk State University Journal of Information Technologies. 2014. Vol. 12. No. 4. Pp. 112-120. ISSN1818-7900. (In Russian)
15. Akoush S., Sohan R., Rice A., Moore A. W., Hopper A. Predicting the performance ofvirtual machine migration. 18th IEEE/ ACM International Symposium on Modelling, Analysis & Simulation of Computer and Telecommunication Systems, MASCOTS2010 (Miami, Florida, USA, 2010). IEEE, 2010. Pp. 37- 46. DOI: 10.1109/ MASCOTS.2010.13
16. Jo C., Cho Y., Egger B. A machine learning approach to live migration modeling. Proceedings of the 2017 Symposium on Cloud
Computing - SoCC'17 (Santa Clara, CA, USA, September 24-27, 2017). New York, 2017. Pp. 351-364. DOI:10.1145/3127479.3129262
17. Vorozhtsov A.S., Tutova N. V. Algorithm for solving optimization problems of resource allocation of data centers in the Internet. T-Comm. 2009. No. S2. Pp. 144-146. (In Russian)
18. Tutov A.V., Tutova N. V., Vorozhtsov A. S. Modeling of resource allocation in cloud data centers. T-Comm. 2017. Vol. 11. No.4. Pp. 76-80.
19. Buyya R., Ranjan R., Calheiros R. N. Modeling and Simulation of Scalable Cloud Computing Environments and the CloudSim Toolkit: Challenges and Opportunities. Proc. of International Conference on High Performance Computing & Simulation (Leipzig, 21-24 June 2009). IEEE, 2009. Pp. 1-11. DOI: 10.1002/spe.995.
20. Clark C., Fraser K., Hand S., Hansen J. G., Jul E., Limpach C., Pratt I., Warfield A. Live migration of virtual machines. Proceedings of the 2nd Symposium on Networked Systems Design & Implementation, NSDI'05 (Berkeley, CA, USA May 02-04, 2005). USENIX Association, 2005. Vol. 30. No. 4. Pp. 273-286.
21. Vorozhtsov A.V., Toutova N. V., Toutov A. V. Resource control system stability of mobile data centers. 2018 Systems of Signals Generating and Processing in the Field of on Board Communications (Moscow 14-15March 2018). IEEE, 2018. Pp. 1-4.
22. Tutov A. V. Resource control system of cloud datacenters. Datchiki & Systemi [Sensors & Systems]. 2018. No.7. Pp.15-20. (In Russian)
23. Legkov K. E. (2016). Models and methods of monitoring parameters characterizing the state of the infocommunication systems a special purpose. T-Comm. Vol. 10. No. 1. Pp. 11-18. (In Russian)
24. Baginyan A., Dolbilov A., Korenkov V. Equal cost multi pathing in high power systems with TRILL. T-Comm. 2017. Vol. 11. No. 4. Pp. 14-19. (In Russian)
25. Baginyan A., Dolbilov A., Korenkov V. Equal cost multi pathing in high power systems with TRILL. T-Comm. 2017. Vol. 11. No. 4. Pp. 14-19. (In Russian)
26. Dokuchaev VA., Kalfa A.A., Mytenkov S. S., Shve-dov A. V. Technical solutions analysis for the modern Data Centers. T-Comm. 2017. Vol. 11. No. 6. Pp. 16-24. (In Russian)
27. Goncharov A. M., Chashhin S. V., Prokhorov M.A. Approach to the solution of the problem of estimation of steady functioning of the information system on the example of data-processing centre. T-Comm. 2017. Vol. 11. No. 4. Pp. 20-25. (In Russian)
28. Vorozhtsov A. S., Tutova N. V., Tutov A. V. The technique of optimal virtual server placement in data centers. T-Comm. 2015. Vol. 9. No. 7. Pp. 5-10. (In Russian)
29. Abaev P. O., Razumchik R. V., Uglov I. V. Analysis of the SIP-traffic model of a contact center according to the treatment results of network measurements. T-Comm. 2013. Vol. 7. No. 11. Pp. 4-10. (In Russian)
30. Gudkova I.A., Maslovskaya N. D. A probable model for analysis of delays of access to the infrastructure of cloud calculations with a monitoring system. T-Comm. 2014. Vol. 8. No. 6. Pp. 13-15. (In Russian)
Vol 10 No 6-2018, H&ES RESEARCH = PUBLICATIONS IN ENGLISH: INFORMATICS, COMPUTER ENGINEERING AND CONTROL
МОДЕЛИ И МЕТОДЫ РАСПРЕДЕЛЕНИЯ РЕСУРСОВ ИНФОКОММУНИКАЦИОННОЙ СИСТЕМЫ ОБЛАЧНЫХ ЦЕНТРОВ ОБРАБОТКИ ДАННЫХ
ТУТОВ АНДРЕИ ВЛАДИМИРОВИЧ,
г. Москва, Россия, andrew_vidnoe@mail.ru
КЛЮЧЕВЫЕ СЛОВА: управление ресурсами; облачные технологии; размещение виртуальных машин; живая миграция; центр обработки данных.
АННОТАЦИЯ
Основной проблемой современных центров обработки данных является потребление колоссальных объемов электроэнергии, значительная часть которой идет на инфокоммуникационную систему и систему охлаждения, которая начинает усиленно работать в случае неравномерного распределения температуры в зале. Поэтому необходимо совершенствовать методы распределения ресурсов для минимизации энергопотребления, при этом обеспечивая высокий уровень качества сервисов. Перераспределение ресурсов в облачном центре обработки данных происходит путем живой миграции виртуальных машин, которая дополнительно нагружает систему и мешает осуществлять мониторинг серверов. Данный факт учитывать в процессе управления. В настоящее время существует большое количество работ, посвященных отдельным вопросам оптимального распределения и управления ресурсами облачных центров обработки данных. Однако в известных работах отсутствует полный цикл работ. В данной работе предложены модели и методы полного цикла работ по оптимизации и управлению ресурсами инфокоммуникационной системы облачного центра обработки данных. В частности, предложена модель первоначального размещения виртуальных машин в виде задачи многокритериальной оптимизации и метод ее решения. Приведена двухуров-
невая система управления ресурсами, которая включает в себя локальные и глобальный контроллер. Локальный контроллер осуществляет мониторинг загрузки и температуры серверов и делает прогноз на следующее окно наблюдения. Для прогнозирования предложено использовать метод группового учета аргументов. Для определения размера окна наблюдения необходимо учитывать длительность миграции. Проведено исследование двух видов живой миграции и предложен метод расчета длительности живой миграции, на основе нахождения аналитического выражения плотности вероятности, позволяющего с определенной вероятностью определить критерий закрытия окна на локальных контроллерах в системе управления ресурсами центров обработки данных. Для сервисов SaaS и PaaS, использующих горизонтальное масштабирование, предложена модель двухкритериальной оптимизации числа виртуальных машин в кластерах крупного многозвенного приложения, которую предложено решать комбинированным методом последовательных уступок и ограничений.
СВЕДЕНИЯ ОБ АВТОРЕ:
Тутов A.B., старший преподаватель Московского технического университета связи и информатики.
Для цитирования: Тутов A.B. Модели и методы распределения ресурсов инфокоммуникационной системы облачных центров обработки данных// Наукоемкие технологии в космических исследованиях Земли. 2018. Т. 10. № 6. С. 100-107. doi: 10.24411/2409-54192018-10192 (Англ.)
