With scientific computing in the cloudgaining popularity and using every time larger datasets, high performance storage I/O in virtualizedenvironments is substantially increasing in importance.However, exploiting the performance potential ofthe storage I/O on today’s virtualized architecturesis complex, due to the limitations of POSIX standardfor storage I/O and the lack of integration of relatedmechanisms such as data sharing, storage I/O coordination,relaxing the consistency semantics, and datalocality awareness.In this paper we propose VIDAS (Virtualized DAtaSharing), an object-based virtualized data store thattargets to integrate the above mechanisms through asimple and powerful interface. VIDAS can be used toefficiently and consistently share access to externallystored data in virtualized environments based on ashared pool of storage objects. We show how VIDAScan be used for straightforwardly implementing I/Ocoordination and data sharing for two common highperformancepatterns: inter-domain write-reader andinter-domain collective I/O. We present the implementationand evaluation of VIDAS for the Xen virtualizationsolution. In addition, we present a novelmechanism for efficiently sharing memory among anarbitrary number of virtual machines.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados