Executing Analytics and Fusion Workloads on Transient Computing Resources in Tactical Environments

Ahmed Ali-Eldin, Deepak Ganesan, Heesung Kwon, Benjamin M. Marlin, Prashant J. Shenoy, Mani B. Srivastava, Don Towsley

Abstract: Transient computing is an emerging paradigm where the availability of resources is uncertain, transient and varies over time. Transiency arises in many environments such as cloud computing, ad-hoc computing, and energy-aware computing and raises many new challenges for applications. In a cyber-physical system, for instance, transiency can cause the availability and capacity of computing, network and sensing resources to vary over time in an unpredictable manner. At the same time, the data produced by sensors in the cyber-physical environment will need to be processed by fusion and analytic tasks that themselves have time-varying resource needs. In this paper, we argue for the use of the transient computing paradigm as a principled approach for handling the challenges seen in highly-dynamic IoT environments. We discuss several challenges in running fusion and analytic workloads in these environments and present initial research directions for addressing these challenges.