تخصیص تکنیک های منابع بهینه و غیر بهینه در مراکز داده محاسبات ابر Optimal and suboptimal resource allocation techniques in cloud computing data centers

Introduction The appeal of cloud computing for clients comes from the promise of transforming computing infrastructure into a commodity or a service that organizations pay for exactly as much as they use. This idea is an IT corporation executive’s dream. As Gartner analyst Daryl Plummer puts it: “Line-of-business leaders everywhere are bypassing IT departments to get applications from the cloud .. and paying for them like they would a magazine subscription. And when the service is no longer required, they can cancel that subscription with no equipment left unused in the corner” [۱]. The idea that centralized computing over the network is the future, was clear to industry leaders as early as 1997. None other than Steve Jobs said:“I don’t need a hard disk in my computer if I can get to the server faster .. carrying around these non-connected computers is byzantine by comparison” [۱]. This applies as well to organizations purchasing and planning large data centers. However, performance remains the critical factor. If – at any point- doubts are cast over a provider’s ability to deliver the service according to the Service Level Agreements (SLAs) signed, clients will consider moving to other providers. They might even consider going back to the buy-and-maintain model. Providers are under constant pressure to improve performance, offer more diverse resource deployment options, improve service usability, and enhance application portability. A main weapon here is an efficient resource allocation system. As in Fig. 1, in the cloud scenario, clients are able to rent Virtual Machines (VMs) from cloud providers. Providers offer several deployment models where VM configuration differs in computing power, memory, storage capacity and platform just to name a few factors. During the rental period, clients require network capabilities. Clients will have data frequently exchanged between client headquarters (or private clouds) and VMs or between two client VMs. The aim here for a scheduler is to schedule VM reservation requests and connection requests in the fastest possible way while using the data center resources optimally. This task is getting even harder with the emergence of the big data concepts. IBM summarized big data challenges into 4 different dimensions referred to as the 4 Vs: Volume, Velocity, Variety, and Veracity [2]. With most companies owning at least 100 TB of data stored and with 18.6 billion network connections estimated to exist now [2], resource allocation efficiency has never been so important. When faced by the task of designing a resource allocation methodology, many external and internal challenges should be considered. An attempt to summarize these challenges can be found in [3]. External challenges include regulative and geographical challenges as well as client demands related to data warehousing and handling. These limitations result in constraints on the location of the reserved VMs and restrictions to the data location and movements. External challenges also include optimizing the charging model in such a way that generates maximum revenue. Internal challenges discussed in [3] include also data locality issues. The nature of an application in terms of being data intensive should be considered while placing the VMs and scheduling connections related to this application. To achieve these performance and cost objectives, cloud computing providers need a comprehensive resource allocation system that manages both computational and network resources. Such an efficient system would have a major financial impact as excess resources translate directly into revenues. The following sections are organized as follows: a discussion of the related research efforts is introduced in the following section leading to this paper’s contribution. Detailed model description is given in “Model description” section. “Mathematical formulation” section presents the mathematical formulation of the problem. The heuristic methods are presented in “Heuristic solution” section. The suboptimal solution is presented in “Suboptimal solution” section. Results are shown and analyzed in “Results” section. Finally, “Conclusion” section concludes the paper and conveys future work.