HGGF(halo-based galaxy group finder)算法实现了基于暗物质晕的星系找群,在研究宇宙大尺度结构及宇宙的演化等领域中占有至关重要的地位。但由于数据规模的增长,急需对HGGF算法进行优化,以缩短运行时间。经分析,算法的热点部分耗时受...HGGF(halo-based galaxy group finder)算法实现了基于暗物质晕的星系找群,在研究宇宙大尺度结构及宇宙的演化等领域中占有至关重要的地位。但由于数据规模的增长,急需对HGGF算法进行优化,以缩短运行时间。经分析,算法的热点部分耗时受到非规则访存的严重影响,因此针对算法的结构和非规则访存模型,提出了数据预排序方法,并分析了该方法如何影响访存过程。在此基础上,利用数据对齐、循环分解进一步优化访存效率,利用负载均衡和互斥变量私有化的方法提高了Open MP的并行效率,最终将HGGF应用使用12线程加速11.6倍,同时取得了更好的可扩展性。主要有三点贡献:(1)分析了HGGF算法的非规则访存问题;(2)提出并分析了数据预排序方法;(3)使用数据对齐、循环分解、负载均衡、互斥变量私有化方法提高了HGGF应用的并行性能。展开更多
In order to resolve the hidden and exposed terminal problems and improve the probability of concurrent packet transmissions for multihop Mobile Ad Hoc Networks （MANETs）, a novel slotted Asyrmaetric Dual-Channel Medi...In order to resolve the hidden and exposed terminal problems and improve the probability of concurrent packet transmissions for multihop Mobile Ad Hoc Networks （MANETs）, a novel slotted Asyrmaetric Dual-Channel Medium Access Control （ADC-MAC） protocol is proposed. It exploits sirmltaneous reservation with less collisions and conision-flee data packet transmissions, and achieves optimal transmission balance on the Control Channel （CCH） and Data Channel （DCH） by adjusting the relationship between Reservation Slot （RS） on the CCH and the data packet Transmission Slot （TS） on the DCH. Transmission interferences can be avoided by only observing CCH for the transmission time of a data packet. The proposed RS and contention micro-slot backoff mechanisms also greatly improve channel access efficiency. Simulation results show that compared to IFEE 802. 11 DCF and -Mc protocols, the proposed protocol can achieve a throughput gain of 88% in singlehop networks and 151% in nltihop networks at the same total data rate.展开更多
New non-volatile memory （e.g., phase-change memory） provides fast access, large capacity, byteaddressability, and non-volatility features. These features, fast-byte-persistency, will bring new opportunities to fault...New non-volatile memory （e.g., phase-change memory） provides fast access, large capacity, byteaddressability, and non-volatility features. These features, fast-byte-persistency, will bring new opportunities to fault tolerance. We propose a fine-grained checkpoint based on non-volatile memory. We extend the current virtual memory manager to manage non-volatile memory, and design a persistent heap with support for fast allocation and checkpointing of persistent objects. To achieve a fine-grained checkpoint, we scatter objects across virtual pages and rely on hardware page-protection to monitor the modifications. In our system, two objects in different virtual pages may reside on the same physical page. Modifying one object would not interfere with the other object. This allows us to monitor and checkpoint objects smaller than 4096 bytes in a fine-grained way. Compared with previous page-grained based checkpoint mechanisms, our new checkpoint method can greatly reduce the data copied at checkpoint time and better leverage the limited bandwidth of non-volatile memory.展开更多
The launch of a Sino-African joint media interview program in Dar es Salaam,Tanzania,on November 28.The initiative aims to help both Chinese and Africans know more about Chinese companies’operations in Africa,said Su...The launch of a Sino-African joint media interview program in Dar es Salaam,Tanzania,on November 28.The initiative aims to help both Chinese and Africans know more about Chinese companies’operations in Africa,said Sun Zhijun（third right）,展开更多
文摘HGGF(halo-based galaxy group finder)算法实现了基于暗物质晕的星系找群,在研究宇宙大尺度结构及宇宙的演化等领域中占有至关重要的地位。但由于数据规模的增长,急需对HGGF算法进行优化,以缩短运行时间。经分析,算法的热点部分耗时受到非规则访存的严重影响,因此针对算法的结构和非规则访存模型,提出了数据预排序方法,并分析了该方法如何影响访存过程。在此基础上,利用数据对齐、循环分解进一步优化访存效率,利用负载均衡和互斥变量私有化的方法提高了Open MP的并行效率,最终将HGGF应用使用12线程加速11.6倍,同时取得了更好的可扩展性。主要有三点贡献:(1)分析了HGGF算法的非规则访存问题;(2)提出并分析了数据预排序方法;(3)使用数据对齐、循环分解、负载均衡、互斥变量私有化方法提高了HGGF应用的并行性能。
基金Acknowledgements This work was supported partially by the National Natural Science Foundation of China under Gants No. 60872011, No. 61171074 the National S＆T Major Project of China under Gant No. 2010ZX03003-003-03+1 种基金 the Program for New Century Excellent Talents in University the Fundamental Research Funds for the Central Universities.
文摘In order to resolve the hidden and exposed terminal problems and improve the probability of concurrent packet transmissions for multihop Mobile Ad Hoc Networks （MANETs）, a novel slotted Asyrmaetric Dual-Channel Medium Access Control （ADC-MAC） protocol is proposed. It exploits sirmltaneous reservation with less collisions and conision-flee data packet transmissions, and achieves optimal transmission balance on the Control Channel （CCH） and Data Channel （DCH） by adjusting the relationship between Reservation Slot （RS） on the CCH and the data packet Transmission Slot （TS） on the DCH. Transmission interferences can be avoided by only observing CCH for the transmission time of a data packet. The proposed RS and contention micro-slot backoff mechanisms also greatly improve channel access efficiency. Simulation results show that compared to IFEE 802. 11 DCF and -Mc protocols, the proposed protocol can achieve a throughput gain of 88% in singlehop networks and 151% in nltihop networks at the same total data rate.
基金Project supported by the National High-Tech R＆D Program （863） of China （Nos. 2012AA01A301, 2012AA010901, 2012AA010303, and 2015AA01A301）, the Program for.New Century Excellent Talents in University, the National Natural Science Foundation of China （Nos. 61272142, 61402492, 61402486, 61379146, and 61272483）, the Laboratory Pre-research Fund （No. 9140C810106150C81001）, and the State Key Laboratory of High-End Server ＆ Storage Technology （No. 2014HSSA01）
文摘New non-volatile memory （e.g., phase-change memory） provides fast access, large capacity, byteaddressability, and non-volatility features. These features, fast-byte-persistency, will bring new opportunities to fault tolerance. We propose a fine-grained checkpoint based on non-volatile memory. We extend the current virtual memory manager to manage non-volatile memory, and design a persistent heap with support for fast allocation and checkpointing of persistent objects. To achieve a fine-grained checkpoint, we scatter objects across virtual pages and rely on hardware page-protection to monitor the modifications. In our system, two objects in different virtual pages may reside on the same physical page. Modifying one object would not interfere with the other object. This allows us to monitor and checkpoint objects smaller than 4096 bytes in a fine-grained way. Compared with previous page-grained based checkpoint mechanisms, our new checkpoint method can greatly reduce the data copied at checkpoint time and better leverage the limited bandwidth of non-volatile memory.
文摘The launch of a Sino-African joint media interview program in Dar es Salaam,Tanzania,on November 28.The initiative aims to help both Chinese and Africans know more about Chinese companies’operations in Africa,said Sun Zhijun（third right）,
