Big data analytics is emerging as one kind of the most important workloads in modern data centers. Hence,it is of great interest to identify the method of achieving the best performance for big data analytics workload...Big data analytics is emerging as one kind of the most important workloads in modern data centers. Hence,it is of great interest to identify the method of achieving the best performance for big data analytics workloads running on state-of-the-art SMT( simultaneous multithreading) processors,which needs comprehensive understanding to workload characteristics. This paper chooses the Spark workloads as the representative big data analytics workloads and performs comprehensive measurements on the POWER8 platform,which supports a wide range of multithreading. The research finds that the thread assignment policy and cache contention have significant impacts on application performance. In order to identify the potential optimization method from the experiment results,this study performs micro-architecture level characterizations by means of hardware performance counters and gives implications accordingly.展开更多
Programs take on changing behavior at nmtime in a simultaneous multithreading （SMT） environment. How reasonably common resources are distributed among the threads significantly determines the throughput and fairness...Programs take on changing behavior at nmtime in a simultaneous multithreading （SMT） environment. How reasonably common resources are distributed among the threads significantly determines the throughput and fairness performance in SMT processors. Existing resource distribution methods either mainly rely on the front-end fetch policy, or make distribution decisions according to the limited information from the pipeline. It is difficult for them to efficiently catch the various resource requirements of the threads. This work presents a spatially triggered dissipative resource distribution （SDRD） policy for SMT processors, its two parts, the self-organization mechanism that is driven by the real-time instructions per cycle （IPC） performance and the introduction of chaos that tries to control the diversity Of trial resource distributions, work together to supply sustaining resource distribution optimization for changing program behavior. Simulation results show that SDRD with fine-grained diversity controlling is more effective than that with a coarse-grained one. And SDRD benefits much from its two well-coordinated parts, providing potential fairness gains as well as good throughput gains. Meanings and settings of important SDRD parameters are also discussed.展开更多
Data cube computation is an important problem in the field of data warehousing and OLAP （online analytical processing）. Although it has been studied extensively in the past, most of its algorithms are designed witho...Data cube computation is an important problem in the field of data warehousing and OLAP （online analytical processing）. Although it has been studied extensively in the past, most of its algorithms are designed without considering CPU and cache behavior. In this paper, we first propose a cache-conscious cubing approach called CC-Cubing to efficiently compute data cubes on a modern processor. This method can enhance CPU and cache performances. It adopts an integrated depth-first and breadth-first partitioning order and partitions multiple dimensions simultaneously. The partitioning scheme improves the data spatial locality and increases the utilization of cache lines. Software prefetching techniques are then applied in the sorting phase to hide the expensive cache misses associated with data scans. In addition, a cache-aware method is used in CC-Cubing to switch the sort algorithm dynamically. Our performance study shows that CC-Cubing outperforms BUC, Star-Cubing and MM-Cubing in most cases. Then, in order to fully utilize an SMT （simultaneous multithreading） processor, we present a thread-based CC-Cubing-SMT method. This parallel method provides an improvement up to 27% for the single-threaded CC-Cubing algorithm.展开更多
基金Supported by the National High Technology Research and Development Program of China(No.2015AA015308)the State Key Development Program for Basic Research of China(No.2014CB340402)
文摘Big data analytics is emerging as one kind of the most important workloads in modern data centers. Hence,it is of great interest to identify the method of achieving the best performance for big data analytics workloads running on state-of-the-art SMT( simultaneous multithreading) processors,which needs comprehensive understanding to workload characteristics. This paper chooses the Spark workloads as the representative big data analytics workloads and performs comprehensive measurements on the POWER8 platform,which supports a wide range of multithreading. The research finds that the thread assignment policy and cache contention have significant impacts on application performance. In order to identify the potential optimization method from the experiment results,this study performs micro-architecture level characterizations by means of hardware performance counters and gives implications accordingly.
基金the Hi-Tech Research and Development Pro-gram (863) of China (No. 2006AA01Z431) the Key Science andTechnology Program of Zhejiang Province (Nos. 2007C11068 and2007C11088), China
文摘Programs take on changing behavior at nmtime in a simultaneous multithreading （SMT） environment. How reasonably common resources are distributed among the threads significantly determines the throughput and fairness performance in SMT processors. Existing resource distribution methods either mainly rely on the front-end fetch policy, or make distribution decisions according to the limited information from the pipeline. It is difficult for them to efficiently catch the various resource requirements of the threads. This work presents a spatially triggered dissipative resource distribution （SDRD） policy for SMT processors, its two parts, the self-organization mechanism that is driven by the real-time instructions per cycle （IPC） performance and the introduction of chaos that tries to control the diversity Of trial resource distributions, work together to supply sustaining resource distribution optimization for changing program behavior. Simulation results show that SDRD with fine-grained diversity controlling is more effective than that with a coarse-grained one. And SDRD benefits much from its two well-coordinated parts, providing potential fairness gains as well as good throughput gains. Meanings and settings of important SDRD parameters are also discussed.
基金supported in part by a grant from HP Labs China,the National Natural Science Foundation of China under GrantNo.60496325the Main Memory OLAP Servers Project
文摘Data cube computation is an important problem in the field of data warehousing and OLAP （online analytical processing）. Although it has been studied extensively in the past, most of its algorithms are designed without considering CPU and cache behavior. In this paper, we first propose a cache-conscious cubing approach called CC-Cubing to efficiently compute data cubes on a modern processor. This method can enhance CPU and cache performances. It adopts an integrated depth-first and breadth-first partitioning order and partitions multiple dimensions simultaneously. The partitioning scheme improves the data spatial locality and increases the utilization of cache lines. Software prefetching techniques are then applied in the sorting phase to hide the expensive cache misses associated with data scans. In addition, a cache-aware method is used in CC-Cubing to switch the sort algorithm dynamically. Our performance study shows that CC-Cubing outperforms BUC, Star-Cubing and MM-Cubing in most cases. Then, in order to fully utilize an SMT （simultaneous multithreading） processor, we present a thread-based CC-Cubing-SMT method. This parallel method provides an improvement up to 27% for the single-threaded CC-Cubing algorithm.
