Autonomic Multi-target Deployment of Science and Engineering HPC Applications

Switching between computational resources in order to select the most suitable computational environment for application execution is a desirable component of utility like computing. As IaaS clouds offer bare-bones computational platforms, they seem to be perfect choice for any application, however, heterogeneity in operating systems and vendors´ policies wrt system images hamper use of runtime environments from different providers. In the context of scientific and engineering applications, flexible execution is further obstructed by hardware incompatibility and inability to use automating provisioning tools as this software should be tuned for specific targets to achieve high performance and usually is distributed in sources with atypical requirements. We offer ADAPT, a novel tool that concentrates on provisioning of scientific and engineering software and addresses its extraordinary needs. Our method is an autonomic execution monitor that exploits feedback from the runtime and resolves missing dependencies. ADAPT investigates dynamic self deployment of applications in science and engineering and this automation facilitates migration between unspecialized resources such as IaaS clouds, clusters, and workstations. As a result, improved flexibility in software execution makes utility-like computing pragmatic even for hard-to-deploy applications.