Melhorando a eficiência de sistemas em nuvem através de coalocação de processos sensíveis à latência e de melhor esforço

Abstract

Cloud computing offers a centralized and service-oriented infrastructure for users worldwide. A cloud environment allows for the execution of critical user interaction-sensitive applications(such as online web searches). And it is also used to run best-effort batch-oriented applications, such as audio/video compression and web page indexing. Data centers are designed on a large scale to support the various cloud services distributed across hundreds of thousands of servers, with significant operational and capital expenses. Hence, optimizing the use of server resources can result in substantial cost savings in these systems. The possibility of exploring idle moments of servers that perform critical services, especially during periods of low demand, presents itself as an encouraging research opportunity. Therefore, this research aims to allow the co-alocation of different cloud application processes (latency-sensitive and best-effort) on the same server to increase utilization, thus, improving efficiency. The study demonstrates that co-locating processes, being latency-sensitive or best-effort, deliver greater efficiency of cloud servers, provided that the performance of critical application processes is guaranteed. The study explored using specific schedulers for each type of process. Latency-sensitive ones were scheduled and parameterized via real-time scheduling. Batch application processes used fair best-effort scheduling. The proposed scheduler, EDFCoaloc, showed promising results, up to 150% better than the standard operating system scheduler.