Support for Virtual Machine Malleability in Cloud Computing Using Migratable Components

Cloud computing is an emerging distributed computing paradigm that promises to offer cost-effective scalable on-demand services to users, without the need for large up-front infrastructure investments [1]. A key enabling technology for cloud computing is virtualization. With virtualization, service providers can ensure isolation of multiple user workloads, provision resources in a costeffective manner by consolidating virtual machines (VMs) onto fewer physical resources when system load is low, and quickly scale up workloads to more physical resources in peak periods. Cloud computing users also benefit from VM abstraction in that they can quickly deploy VM images with preloaded operating systems, software applications, and application data onto cloud environments without being concerned with lower-level physical infrastructure details. As cloud computing brings significant benefits to service providers and users, it also introduces new challenges. Its pay-per-use business model gives service users such flexibility in purchasing a service that they could try a service with lower cost and turn away from a service whenever they like, which results in fluctuating user workloads. Meanwhile, the available physical resources for one VM image are non-dedicated and dynamic, because service providers usually multiplex multiple VMs onto one single physical machine (PM) and would turn on/off PMs from time to time for the sake of maintenance or energy savings. In such a computing environment, VMs are likely to be either overprovisioned, incurring unnecessary cost, or underprovisioned, which may degrade the quality of service. Thus, in order for the cloud computing paradigm to fulfill its promises, it is necessary to investigate adaptive VM management techniques that enable the VMs to achieve optimal quality of service in spite of fluctuating user workloads and available physical resources. While existing works on adaptive VM management mainly focus on VM migration [2, 3, 4], VM malleability, which is the ability to change the gran-