Concurrent Adaptive Computing for Heterogeneous Environments (CACHE)

Selected users in a military tactical network environment have computing needs that may not be satisfied by a single resource. The majority of these resources are underutilized when executing non-demanding applications and may be exhaustible concurrently by a user with higher computing needs. This dissertation presents a framework for aggregating these underutilized resources into a single coherent efficient system for concurrent workload execution. Furthermore, the framework only aggregates the resources needed and releases them upon completion. Upon aggregation, the framework also allows for an adaptation of the workload to meet a time constraint.;This framework takes into consideration the different performance capabilities of the aggregated resources. These variant capabilities add significant noise to evaluating and representing the power in the system. The variance among resources in the system is unappealing since it causes unpredictable performance and lack of control of a distributed system composed of these resources.;Membership into the aggregated system is done by selecting the best suited resources for the task to execute. Essentially, it entails the "assembling of the all-stars" from the available resources by finding the subset of the most invariant, high performing resources to execute the workload within a specified time interval. Therefore, a homogenization of the aggregated system reduces the variability of resource capabilities which results in a more controlled and predictable system.;The second contribution of this dissertation is a workload adaptation algorithm which generates an answer within a specific time interval for an already established efficient aggregated system. This adaptation produces an answer within a specified time interval. If the original workload is unable to be completed in the time given even with the aggregation running at peak performance, then the proposed framework adapts the original workload by changing the size of the problem. This change of the workload size guarantees an answer within the allocated time, also known as the "70% solution".