CC BY
26
THE NON-MARKOVIAN MODEL OF VIRTUAL DESKTOP CLOUD SERVICES TERMINAL SESSION
Almaz A. Suleymanov,
Moscow Technical University of Communications and Informatics, Moscow, Russia, efquantum@gmail.com
Keywords: cloud services, virtual desktop, server, terminal session, response time.
Cloud technologies became one of the fastest growing trend in the infocomm environment. With the development of various schemes, they received wide popularity among users working with all types of applications. Therefore, one of the most actual issues is the question of ensuring an acceptable quality of service.
Consider a "Virtual Desktop", or DaaS (Desktop-as-a-Service). The main idea of this service is to provide the user a full workplace "from the cloud" on any user device that has the appropriate specific cloud software agent platform. The user in this case would work just as well as on a normal workplace, but getting extra benefits of cloud: around the clock availability of the desktop, the absence of binding to the office, protection against data loss.
It seems appropriate to divide service's work process into two phases for the convenience of description and simulation: the phase of the initialization phase of a terminal session and the terminal session, which is considered in this paper. One of the main parameters of the system, serving requests from users is a response time because the user perceived quality of service directly depends on it. Proposed a non-Markovian analytical model of "virtual desktop" service's terminal session operation process, based on an approximate method that takes into account the first two moments. Derived the expressions for the average response time, as well as its depending on the variation coefficients. Estimated the average response time assuming the initial data close to real.
Information about author:
Almaz A. Suleymanov, Moscow Technical University of Communications and Informatics, postgraduate, Moscow, Russia.
Для цитирования:
Сулейманов А.А. Немарковская модель терминальной сессии облачной услуги "виртуальный рабочий стол" // Т-Сотт: Телекоммуникации и транспорт. 2017. Том 1 1. №4. С. 72-75.
For citation:
Suleymanov A.A. (2017). The non-markovian model of virtual desktop cloud service's terminal session. T-Comm, vol. 1 1, no.4, pp. 72-75.
Organization of services "from the cloud" have become a popular and common way of their provision. Cloud services have been applied in different areas: a corporate environment, the public sector, industry, media sector due to their convenience, af-fordability and safety of data.
Among the cloud-based services due to the specifics of its technological infrastructure the "virtual desktop" service (eng. Desktop as a Service, DaaS) occupies a special place, basic idea of which is described in [1]. It can be stated as providing the full work place "from the cloud" through the telecommunication network to the user device.
Service's infrastructure includes the following main elements:
• on the provider side — hardware and software systems (server cloud platform, the virtual izati on system, data storage systems, the virtual switch), network equipment (network routers and switches);
• on the user side — terminal equipment with installed software agent.
Communication between the user device and the server is carried over the data network by remote desktops delivery protocols.
User devices may be personal computers, mobile devices (smartphones, tablets), as well as thin clients (quietest compact terminal devices placed in the workplace, the main function of which is the connection of user to the server service over the network). An architecture of service, implementation schemes, the provisions of ITU-T Recommendations dedicated to service more detailed systematized in [2].
After analyzing the features of the cloud-based services, in particular, "virtual desktop" service's, we can conclude that the response time is one of the key characteristics of the system, therefore, it directly affects the quality of service as a whole. The response time is the time that elapses from the moment of pressing the keyboard or mouse movement to the result's appearance on the user's monitor - the server response to this action. This time is the sum of the service time of the server Ts, double-sided transport network delay of data transmission Ttr and processing response time on the user device (rendering time) Tv:
T = TS+2T„+TV. (1)
More detailed analysis of these time slots is given in [3J.
The factors influencing the quality of service (QoS) from the point of view of the user are formulated in [4]. For this purpose, the information transmitted is divided into three categories: audio, video, data, for each of that is shown, in particular, the values of the permissible delay. The valid value of transport delays for online interaction according to this Recommendation is 250 ms. In a study [5] shown that the value of the transport delay for a comfortable remote desktop's user experience should not exceed 150 ms.
It is appropriate to divide the service's working process in two phases, based on the logic of the service. In the first phase the initialization of a terminal session takes place (connection, authorization and authentication processes). The approach to the development of the first phase's analytical model was shown in [6, 7].
In the second phase connected to a terminal session users work with their desktops. After each user action from the terminal device to the server a request containing the mouse movement or keyboard keys command is sent across the network. In
COMPUTER SCIENCE
response to request the server sends in the opposite direction the desktop image including changes caused by user action. In other words, there is an interactive exchange of data via a data network, which can be either a local (LAN), or global (WAN),
The aim of this study is to determine the average response time of the "virtual desktop" service in the phase of terminal session.
The issue of work is the second phase of the service: Non-Markov model of terminal session is proposed. It is made in accordance with the approach based on an approximate method that takes into account the first two points. This approach allows us to consider different types of service time distributions at the nodes and the distance between adjacent applications in the stream distributions.
Service's second phase can be mathematically described as an open queuing network. It's nodes simulate the operation of the elements of services architecture. In accordance with the method [8, 9], each node in the network can be represented as queue system type G / G / 1. We will take into account the first two points (mean value and dispersion) of a random variables describing the distance between adjacent applications in flow and duration of service in nodes. We will not consider the dispersion in pure form, but will use the coefficients of variation. The graph of such a network is shown in Figure 1.
x3 V2 f1
- 3K 2K- .
/ A1
In the graph nodes are identified as follows: I - the agent of the virtual machine; 2 - the user device agent; 3, 4 - the core processors of the user device. Assume that the nodes 3 and 4 are identical, i.e. have the same service parameters. In practice, this case is often distributed and corresponds to the basic case of a multi-core processor use.
The following notation introduced:
Py - the probability of request's transition from node i to node j\ N - number of nodes; - the intensity of the ilow of requests from outside into node i; An - the intensity of the flow of requests in node i; Mtp; = 1/X,,; - the average distance between the requests (mean value of the distance between the requests); cA(i) — the variation coefficient of the distribution of distances between the corresponding requests in the flow; Cn(i) - the variation coefficient of the distribution of the service time in node /; fii - the intensity of the service applications in the node /; hi = l/|Ti - average service time in the node (mean value of service time); T* - average response time without taking into account the transport delay; T - the overall average response time.
Intensities of request streams to the node i are determined by the equilibrium equation system: Xi = A^ + ¿jPji = l-N-
To determine the response time will follow the following algorithm of actions:
1, Define ML.j,j on the basis of the intensities of incoming flows in nodes.
2. Calculate the variation coefficients of the serv ice time distribution in the nodes and the variation coefficients of distances between adjacent requests in the streams.
3. For each node calculate the average waiting time W; that is in the sum with hj gives the average time spent by the request in the node. Adding residence times of requests spent in nodes lying on the route of requests, we obtain T of a network.
As shown above, the average service time in the node is determined from the expression h, = 1/p.j. The average waiting lime in the node according to [8, 9] is given by Kramer and Langen-bach-Belz:
P,
*>, = h, ■
where
2(1-A)
• [(C,(O)2 HcB(0)2]■ zip,,(cA(2)
exp
exp
-2(1-A)
3A
-(I-Pi)
MO)' +(e.C0)J
Jfc,(/)<1.
.if »„(Oil.
Wherein pj = Xj / fi,. Following request's processing logic on the user device, it follows that the completed request leaves the network through the node 3, or through the node 4. According to the statement of the problem parameters of units 3 and 4 are the same, then the average response time should include time periods (waiting service), corresponding to only three nodes, for example 1, 2 or 3 nodes. Thus, the total waiting time on application nodes 1, 2, 3 is determined by the formula:
Wr = Wj + + VI'3. (3)
The total average service time on these sites: hT =h, +h2 + V (4)
Average response time (without transport delay): T' =wr+hr. (5)
Consider a numerical example, using the values that are close to reality. Lei 2,5 1/c. jii = 10 1/c, |i2 = 35 1/c, na = m= 15 1/c. Taking into account that node2 distributes the requests by nodes 3 and 4 with equal probability, we can define the incoming flows intensity values: = >.2 = 2.5 l/c, ^,3 = ^4= 1.25 1/c. Then: M
Mcp4= 1/^=0.8 %
To calculate the coefficients of variation we use the UDN algorithm given in [8]. It is suitable for network analysis in a wide range of loads, while maintaining relative Simplicity. Write a system of linear algebraic equations and use the data from the statement of the problem. To estimate the coefficients of variation of the service time distribution Ce(i) an experimental study was maintained. Experimental test bench consisted of the server and a hundred users, Wc analyzed the average service time on the server, the average render lime, as well as their standard deviations. Received value of CH( 1) was 1.5, CB(2)- 1.75.
Write a system of linear algebraic equations by UDN algorithm for the network:
- (*X1 -4 )PI = 0 +5jb(MPIPI, i*=Mi (=1 *=l
where yA(k,i) = ^(CA\k,i)-1),
£pi = l/Xo = 0.4 c, Mcp; = 1/ X2 = 0-4 C, Mcp3 = 1/ Xi = 0.8 c,
Coefficient yA(i) relates to the flow into node i; yA(k,i) relates to the flow from node k into node i; yA(0,i) relates to the flow from outside (node 0) into node i; yJi) relates to the service at node i. ] f yA( i > are known, then CA(i) can be calculated by the formula:
e,(0=Ji+
7Á0
I
(6)
Thus, the equation solution boils down to finding gamma coefficients. Write down the system, eliminating the vanishing terms of the sum on the basis of the routing matrix:
yA(\) = yA(0,\\
r,(2)-r,(D(l-prW, - /„(DA2/».!. T.a (3) - Y A (2X1 - p\ )p\i = h (2 )p\p\s>
/a(4) - ya(2){\ - pDpI = yB(2)p;p¡A.
After using numerical values from the statement of the problem and solving the system of equations, we get: yA(\) = yA(2) = 13.125, yA(3) = rA(*) = 13.196. Then find all C4 (z) using formula (6). We obtain: C,(2) = C¿(1) = 2.5, C,(4) = CI(3) = 3.4.
Then we estimate the time characteristics of the network by the formulas (2-4) by calculating them for each node, and then summing by the path of requests following. For these initial data value response time excluding transport delay was 1.1s.
Conclusion
Thus, the analytical model of DaaS service's terminal session proposed. It is based on the non-Markov approach - approximate method, taking into account the first two points. The resulting model allows us to estimate the temporal characteristics of the service in question, in particular, the key one - the response lime.
References
1. ITU-T Recommendation Y.3503 (05/14). Requirements for desktop as a service.
2. Specification of the main functions of Microsoft's RDP ¡electronic resource] - Access: http://msdn.microsoft.com/en-us/library/cc240445( PROT. 10).aspx
3. Suleymanov A.A. The quality of cloud services such as "virtual desktop" . T-Comm. 2015, Vol. 9. №7. pp. 31 -35. (in Russian)
4. ITU-T Recommendation G. 1010. End-user multimedia QoS categories.
5 Dlisí M„ Napolitano S.. Longo S.. Niccolini S. A closer look at Thin-Client connections: Statistical Application Identification for QoE Detection // Communications Magazine IEEE. Vol. 50. Is.11. DOI: 10.1109/MCOM.2012.6353701.
6. Suleymanov A.A. The analytical model of the process of establishing the session of cloud services such as "virtual desktop" // Proceedings of the !lth International Scientitic and Technical Conference "Advanced technologies in communication tools". Vladimir. Vladimir State University. 2015. pp. 271-273. (inRussian)
1. Suleymanov A.A., Netes V.A. Analysis of the connection time to the cloud service "virtual desktop" . T-Comm. 2016. Vol.10. №7. pp. 41-46. (in Russian)
8. Basharin G.P.. Bocharov P.P.. Cogan J.A. Queuing analysis in computer networks. Theory and methods of calculation. Moscow; Science. 1989. 336 p. (in Russian)
9. Shneps M.A. Information distribution systems. Measurement Methods: A Reference Guide. Moscow: Communication. 1979. 344 p. (in Russian)
T-Comm Tom 11. #4-2017
НЕМАРКОВСКАЯ МОДЕЛЬ ТЕРМИНАЛЬНОЙ СЕССИИ ОБЛАЧНОЙ УСЛУГИ
"ВИРТУАЛЬНЫЙ РАБОЧИЙ СТОЛ"
Сулейманов Алмаз Авхатович, Московский Технический Университет Связи и Информатики, Москва, Россия,
efquantum@gmail.com
Дннотация
Облачные технологии давно стали одним из наиболее динамично развивающихся направлением в инфокоммуникационной среде. С развитием различных схем предоставления, они получили широкую популярность среди пользователей, работающих со всевозможными типами приложений. Поэтому одним из актуальных вопросов становится вопрос обеспечения приемлемого качества услуги.
Рассмотрена услуга "виртуальный рабочий стол" или DaaS (Desktop-as-a-Service). Основная идея этой услуги заключается в предоставлении пользователю полноценного рабочего места "из облака" на любое пользовательское устройство, имеющее соответствующий определенной облачной платформе программный агент. Пользователь при этом будет работать точно также, как и на обычном рабочем месте, но получая дополнительные преимущества облака: круглосуточную доступность рабочего стола, отсутствие привязки к офису, защиту от потерь данных.
Для удобства описания и моделирования процесс работы услуги уместно разделить на две фазы: фазу инициализации терминальной сессии и фазу работы терминальной сессии, которая рассмотрена в данной работе. Одним из основных параметров системы, обслуживающей запросы от пользователей, является время отклика, потому, что от него напрямую зависит воспринимаемое пользователем качество услуги.
Приведена аналитическая немарковская модель процесса работы терминальной сессии услуги типа "виртуальный рабочий стол", основанная на приближенном методе, учитывающем первые два момента. Получены выражения для среднего времени отклика, а также его зависимости от коэффициентов вариации. Проведена оценка среднего времени отклика для исходных данных, близких к реальным.
Ключевые слова: облачная услуга, виртуальный рабочий стол, сервер услуги, терминальная сессия, время отклика. Литература
1. ITU-T Recommendation Y.3503 (05/14). Requirements for desktop as a service.
2. Нетес В.А., Сулейманов А.А. Услуга "виртуальный рабочий стол" и особенности ее реализации // Вестник связи. 2016. №9. С. 12-16.
3. Сулейманов А.А. Качество облачных услуг типа "Виртуальный рабочий стол" // T-Comm: Телекоммуникации и транспорт. 2015. Т. 9. №7. С. 31-35.
4. ITU-T Recommendation G. 1010. End-user multimedia QoS categories.
5. Dusi М.,Napolitano S., Longo S., Niccolini S. A closer look at Thin-Client connections: Statistical Application Identification for QoE Detection // Communications Magazine IEEE. Vol. 50. Is. 1 1. DOI: 10.1 I09/MC0M.20I2.635370I.
6. Сулейманов А.А. Аналитическая модель процесса установления сессии облачной услуги типа "Виртуальный рабочий стол" // Материалы I 1-й международной научно-технической конференции "Перспективные технологии в средствах передачи информации". Владимир: ВлГУ, 2015. С. 271-273.
7. Сулейманов А.А., Нетес В.А Анализ времени подключения к облачной услуге "виртуальный рабочий стол" // T-Comm: Телекоммуникации и транспорт. 2016. Т.10. №7. С. 41-46.
8. Башарин Г.П., Бочаров П.П., Коган Я.А. Анализ очередей в вычислительных сетях. Теория и методы расчета. М.: Наука, 1989. 336 с.
9. Шнепс М.А. Системы распределения информации. Методы расчета: Справочное пособие. М.: Связь, 1979. 344 с.
Информация об авторе:
Сулейманов Алмаз Авхатович, Московский Технический Университет Связи и Информатики (ФГОБУ ВПО МТУСИ), аспирант кафедры ССиСК, Москва, Россия.
