| This thesis introduces a novel Low Level Resource Sharing virtualization paradigm. By using a lightweight and generic architecture, this paradigm allows for a dynamic deployment of components providing reconfiguration, adaptability and distributed management. Memory and peripherals, such as graphics co-processors, storage and input/output devices, can be borrowed from underutilized systems when needed by interfacing them at the device (driver) level. The remote virtual resources are seen by the borrowing system as local devices. Using observations on the utilization patterns of workstations' resources, together with observations on technological trends, this thesis proves that sharing resources over tightly coupled networks is possible and can lead to considerable advantages compared to the utilization of local resources. |
