طراحی شبکه مرکز داده با استفاده از فرآیند تحلیل سلسله مراتبی
|کد مقاله||سال انتشار||تعداد صفحات مقاله انگلیسی||ترجمه فارسی|
|6314||2013||10 صفحه PDF||23 صفحه WORD|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Computer Networks, Volume 57, Issue 3, 26 February 2013, Pages 658–667
مروری بر فرایند تحلیل سلسله مراتبی (AHP)
طراحی شبکه مرکز داده بااستفاده از AHP
جمعیت و موقعیت گره
نتایج استفاده از CM1
نتایج استفاده از CM2
مقایسه با روش شمارش
The demand for delivery services for large-sized content such as video has increased dramatically, and the use of cloud computing services in which users can use IT services via networks has also increased. To provide these services with high quality and high reliability, ISPs need to carefully design network topology and the positions of data centers. However, network topology and data center location strongly affect various evaluation criteria, such as cost, path length, and reliability; therefore, these criteria with different respective units need to be considered simultaneously when designing a data center network. The analytic hierarchy process (AHP) is a way to make a rational decision considering multiple criteria. This paper proposes to design data center networks by evaluating both network topology and data center locations simultaneously using AHP and also shows the numerical results of applying the proposed design method to the three areas of Japan, USA, and Europe. We investigate the properties of desirable data center networks in these three areas, and compare the results with those obtained by the enumeration method.
User demand for viewing video over networks is strong, so the number of people using user generated content (UGC) delivery services, e.g., YouTube, has increased dramatically. As the transmission capacity of access links grows, delivery services for rich content of high quality and huge size, e.g., movies and TV dramas, have been widely provided by many ISPs. Moreover, cloud computing services, in which ISPs instead of users own and manage computer hardware, software, and data and provide computing services to users with usage-based charging, have been widely used. The infrastructure providing these distribution services for rich content and cloud computing services consists of data centers with storages and processors of large capacities and networks connecting multiple data centers at geographically distant locations and users. In this paper, we call the infrastructure consisting of data centers and networks a data center network. It is important for ISPs to adequately design data center networks to provide content delivery and cloud computing services of high quality and high reliability at low cost. However, optimally designing data center networks is difficult because the network topology and data center locations of data center networks strongly affect many evaluation criteria such as service quality, reliability, and cost. To increase redundancy to improve reliability, creating more routes between user nodes and data centers by providing more intermediate nodes, links, and data centers is desirable. However, the increase in links or data centers will also increase equipment and operating costs. For users, avoiding congestion at intermediate nodes and having a shorter path length to reduce packet network delay is desirable. If we decrease the number of links or data centers to reduce network cost, the flexibility of path design is degraded so suppressing path length becomes difficult. Therefore, when designing a network topology, we need to consider multiple incompatible criteria with different respective units. The analytic hierarchy process (AHP) is a way to make a rational decision considering multiple criteria  and . Using AHP, we can reflect the relative importance of each criterion in the evaluation result. AHP treats all the related factors as a hierarchical structure and quantifies qualitative factors, such as the importance of each criterion, using paired comparison. Therefore, we have applied AHP to network topology design to consider multiple criteria simultaneously . However, in , we evaluated only the network topologies without considering the locations of data centers. The locations of data centers as well as the network topology strongly affect service quality, reliability, and cost, so we need to evaluate both the data center locations and network topology when designing data center networks. Therefore, in this paper, we propose to design data center networks by using AHP to consider all the possible combinations of data center locations and network topologies as a candidate set of data center networks.1 We also investigate the results of applying the proposed design method to three areas: Japan, USA, and Europe, and we compare them with those obtained by the enumeration method, i.e., a straightforward approach to strictly obtain the ideal candidate with considering just a single criterion as the optimization target. We describe the related works in Section 2. Section 3 summarizes AHP, and we present the design method of data center networks using AHP in Section 4. We describe the numerical results in Section 5 and conclude the paper in Section 6.
نتیجه گیری انگلیسی
This paper presented a design method of data center networks using AHP. In this method, first we generate candidates for data center networks satisfying the requirements that connectivity between all pairs of nodes be maintained at single link failures (SLFs) and that no links are unused by traffic during normal operation and at any SLFs. Next, we evaluate the generated candidates by AHP using two criteria, i.e., the total link cost and the average path length. We showed the results of applying the proposed design method to three areas: Japan, USA, and Europe, and we investigated the properties of desirable data center networks in these three areas. We also compared the candidates obtained by AHP with those obtained by the enumeration method, and we clarified the superiority of the proposed method using AHP against the enumeration method. When placing S data centers among N nodes and generating candidates for data center networks, we need to check the constraints for all the NCS2N(N−1)/2 possible combinations. Therefore, a method of generating candidates for a large-scale network is future work.