Heuristics and Genetic Algorithms for Adaptive Deployments in Large-scale, Real-time Systems

Distributed, real-time and embedded (DRE) applications are often deployed into mission-critical scenarios where money or lives are at stake. Examples of these types of scenarios include search-and-rescue missions, shipboard computing, satellite infrastructure, power grids, and air traffic control. Once deployed, these DRE applications must service important requests continuously and perpetually, but over time, the environment may change. Hardware may fail or degrade in performance. Computers may move to other locations where latency is worse between important parts of the DRE application. Despite these changes in the environment, the DRE application needs to adapt and continue to respond and operate. In the past, developers or system administrators were required to manually update or reboot the DRE application, analyze or guess the best remaining configurations, and move important processing to the best hardware available. Each of these steps are prone to human error that may result in unacceptable configurations of the DRE application. In this paper, we discuss ongoing work in the context of redeployment of adaptive, mission-critical DRE applications via new heuristics and genetic algorithms that approximate the subgraph isomorphic problem, a known NP complete problem, and middleware and tools that use the results of these heuristics to automate the redeployment process. We provide results that show some of the scenarios that result in perfect deployment approximations, and we also motivate future work to address the blind spots in our approximation techniques.