摘要: Task parallelism is difficult to implement in a distributed setting due to machine unreliability and communication latency. HTCondor, an existing distributed computation framework, is insufficient for addressing these shortcomings. In this report, we present a high level abstraction built on top of HTCondor called the Distributed Completion Service (DCS). The DCS uses multiple different methods to improve upon reliability and latency. Experiment results reveal DCS’s ability to scale and demonstrate a reduction in latency when compared to just using HTCondor alone.