Low-impact profiling of streaming, heterogeneous applications

摘要: Computer engineers are continually faced with the task of translating improvements in fabrication process technology (i.e., Moore's Law) into architectures that allow computer scientists to accelerate application performance. As feature-size continues shrink, architects commodity processors designing increasingly more cores on a chip. While additional can operate independently some tasks (e.g. OS and user tasks), many applications see little no improvement from adding processor alone. For applications, heterogeneous systems offer path toward higher Significant performance power gains have been realized by combining specialized (e.g., Field-Programmable Gate Arrays, Graphics Processing Units) general purpose multi-core processors. Heterogeneous need be programmed differently than traditional software. One approach, stream processing, fits these particularly well because segmented memories explicit expression parallelism. Unfortunately, debugging tools support streaming, do not exist. This dissertation presents TimeTrial, measurement system enables optimization streaming profiling deployed system. TimeTrial performs low-impact measurements dedicating computing resources monitoring aggressively compressing traces statistical summaries guided specification queries interest.