Adaptive scheduling of multiprogrammed dynamic-multithreading applications

Modern parallel platforms, such as clouds or servers, are often shared among many different jobs. However, existing programming runtime systems designed and optimized for running a single job, so it is generally hard to directly use them schedule multiple jobs without incurring high overhead inefficiency. In this work, we develop AMCilk (Adaptive Multiprogrammed Cilk), novel system framework, support multiprogrammed workloads. has client-server architecture where users can dynamically submit the system. that runs these while reallocating cores, last-level cache, memory bandwidth according scheduling policy. exposes interface designer, which allows designer easily build policies meeting requirements of various application scenarios performance metrics, transparently (to designers) enforces also enables its in cloud environment other processes may be sharing with AMCilk. scenario, an external scheduler change resource availability seamlessly adapts changes. The primary feature low-overhead responsive preemption mechanism fast reallocation cores between Our empirical evaluation indicates incurs small overheads provides significant benefits on application-specific criteria set 4 practical applications due core mechanism.