VERTICAL ELASTICITY OF RESOURCE MANAGEMENT ON CLOUDS

Abstract

Cloud computing has emerged as a dominant paradigm for efficiently managing and utilizing computing resources. However, the variability and fluctuations in workload experienced by long-running applications on clouds pose challenges in resource management, necessitating elastic resource allocation for dynamic adjustments. Vertical elasticity, particularly crucial for latency-sensitive applications such as stream applications, enables rapid resource adjustments. Thus, this thesis delves into the vertical elasticity in cloud resource management, with a focusing on the context of stream processing. Specifically, we explore novel resource provisioning strategies and framework designs across three distinct isolation granularities: the operating system level (e.g., virtual machines), the process level (e.g., Linux containers), and the thread level (e.g., slots in Apache Flink). Additionally, we consider two primary user expectations: ensuring performance fairness among general long-running applications and optimizing latency for streaming applications. Through innovative approaches and evaluations, this work provides insights into elastic resource management in cloud environments.