The self distributing virtual machine (SDVM): making computer clusters adaptive

Abstract

The Self Distributing Virtual Machine (SDVM) is a middleware concept to form a parallel computing machine consisting of a any set of processing units, such as functional units in a processor or FPGA, processing units in a multiprocessor chip, or computers in a computer cluster. Its structure and functionality is biologically inspired aiming towards forming a combined workforce of independent units (”sites”), each acting on the same set of simple rules. The SDVM supports growing and shrinking the cluster at runtime as well as heterogeneous clusters. It uses the work-stealing principle to dynamically distribute the workload among all sites. The SDVM’s energy management targets the health of all sites by adjusting their power states according to workload and temperature. Dynamic reassignment of the current workload facilitates a new energy policy which focuses on increasing the reliability of each site. This paper presents the structure and the functionality of the SDVM.