A Multiple Dimension Slotting Approach for Virtualized Resource Management

Recent advances in virtualization technologies can be applied to different research areas, such as virtualized resource management in High Performance Computing (HPC) Grid. On the one hand, the HPC Grid hides to users the complexity of the virtual cluster management, but on the other hand, the Virtual Machines (VM) are managed in the execution domain of the node’s resource manager. Hence, local decisions must be done to meet the quality of service required by each VM. Moreover, a user should be able to define a Service Level Agreement (SLA) with strong and soft requirements for a given resource (CPU, disk, network, etc) without detailed knowledge of the hardware configuration of each node. In this work, we present a fine-grain approach for virtualized resource management in HPC nodes. Our approach improves the resource usage in nodes enabled with virtualization by efficiently partitioning and fine-grain management through multiple dimension slotting. We show in preliminary results that the management of single physical resources gives accurate usage information to a local resource manager (LRM). The assignment of resources is improved by adding a virtual partitioning layer for the subsets of physical resources. Relocating and resizing assigned single resources in the LRM scope improves internal migration mechanisms that could potentially reduce the overhead of external migrations.