Jeremic, NikolausParzyjegla, HelgeMühl, GeroRichling, JanHorbach, Matthias2019-03-072019-03-072013978-3-88579-614-5https://dl.gi.de/handle/20.500.12116/20718In virtualized environments, multiple virtual machines (VMs) usually share a common secondary storage system which is, thus, often subject to a broad range of access patterns and different requirements (e.g., regarding performance, capacity, and reliability) imposed by diverse applications running inside the VMs. Moreover, with applications and VMs being added, started, stopped, and removed, access patterns as well as requirements may vary significantly over time. Ideally, the storage system is able to adapt to changing access patterns while considering application requirements at the same time. However, many storage systems only use a static data organization scheme in terms of drive assignment and data layout that is defined at deployment time, and may become disadvantageous or even inappropriate for the current workload. Although some storage systems employ data migration (e.g., dynamic storage tiering), the data layout remains unchanged due to a prohibitive high reorganization overhead. In this paper, we propose a mechanism for a fine-grained data organization adaptation that includes the data layout. This significantly extends the range of feasible adaptions compared to existing systems. Our approach factors application hints and requirements into adaptation decisions and exploits observations of access patterns as well as the state of the page cache to increase its effectiveness. Furthermore, we present a case study showing the benefits of fine-grained adaptations and discuss two options for the integration of the proposed adaptation mechanism into existing virtual machine monitors (VMMs), also known as hypervisors (HVs).enText/Conference Paper1617-5468