Real time resource management and adaptive parallel programming for a cluster of computers: a comparison of different approaches in a computationally intensive environment

Two classic problems that are computationally intensive and show good speedup and scalability when solved in a parallel programming environment are used to test the different resource allocation and management algorithms used with the node intrusion and failure experiment. We divide the adaptive resource allocation experiments into two groups: (i) automatic survivability and scalability (ii) assessment of real-time quality of service (QoS). In the former, we use different algorithms to detect failed programs, host and network resources and idle times, computing an allocation, enactment of an allocation, and restart notification. We also use different techniques to detect dynamic paths that are receiving poor QoS possibly due to overload and to "scale up" such paths via reallocation. In the latter case, we use different fitness functions to classify the connections and the resources available on the nodes and study the effects of these on the overall resource allocation and the eventual speedup.