内存交换操作中频繁的外存访问是制约程序运行性能的一个重要因素.提出一种结合重复数据删除技术的内存交换机制MSDD(Memory Swap with Data Deduplication),通过删除交换页面中的重复页面减少换出操作,从而降低交换操作对程序运行性能...内存交换操作中频繁的外存访问是制约程序运行性能的一个重要因素.提出一种结合重复数据删除技术的内存交换机制MSDD(Memory Swap with Data Deduplication),通过删除交换页面中的重复页面减少换出操作,从而降低交换操作对程序运行性能的影响.针对交换操作延迟敏感和内存紧缺的特点,MSDD机制采用了改进的MD5算法计算页面的数字指纹以限制其计算时延,同时通过哈希表存储删冗元数据以在内存空间占用率与删冗元数据的检索速率两方面取得权衡.在四种典型应用程序上的实验测试结果表明,MSDD机制可以减少内存换出操作达36.73%,提高程序运行性能达15.0%.展开更多
Recently, a multiple symbol differential （MSD） sphere decoding （SD） algorithm for unitary spacetime modulation over quasi-static channel has been proved to achieve the performance of maximumlikelihood （ML） detec...Recently, a multiple symbol differential （MSD） sphere decoding （SD） algorithm for unitary spacetime modulation over quasi-static channel has been proved to achieve the performance of maximumlikelihood （ML） detection with relatively low complexity. However, an error floor occurs if the algorithm is applied over rapid-fading channels. Based on the assumption of continuous fading, a multiple symbol differential automatic sphere decoding （MSDASD） algorithm is developed by incorporating a recursive form of an ML metric into automatic SD （ASD） algorithm. Furthermore, two algorithms, termed as MSD approximate ASD （MSDAASD） and MSD pruning ASD （MSDPASD）, are proposed to reduce computational complexity and the number of comparisons, respectively. Compared with the existing typical algorithms, i.e., multiple symbol differential feedback detection （MS-DFD） and noncoherent sequence detection （NSD）, the performance of the proposed algorithms is much superior to that of MS-DFD and a little inferior to that of NSD, while the complexity is lower than that of MS-DFD in most cases and significantly lower than that of NSD.展开更多
文摘内存交换操作中频繁的外存访问是制约程序运行性能的一个重要因素.提出一种结合重复数据删除技术的内存交换机制MSDD(Memory Swap with Data Deduplication),通过删除交换页面中的重复页面减少换出操作,从而降低交换操作对程序运行性能的影响.针对交换操作延迟敏感和内存紧缺的特点,MSDD机制采用了改进的MD5算法计算页面的数字指纹以限制其计算时延,同时通过哈希表存储删冗元数据以在内存空间占用率与删冗元数据的检索速率两方面取得权衡.在四种典型应用程序上的实验测试结果表明,MSDD机制可以减少内存换出操作达36.73%,提高程序运行性能达15.0%.
基金Supported by the National Basic Research Program of China (973 Program) (Grant No. 2009CB320403)the National Defense Pre-researchProject of the 11th Five-Year-Plan of China (Grant No. 1060741001020102)
文摘Recently, a multiple symbol differential （MSD） sphere decoding （SD） algorithm for unitary spacetime modulation over quasi-static channel has been proved to achieve the performance of maximumlikelihood （ML） detection with relatively low complexity. However, an error floor occurs if the algorithm is applied over rapid-fading channels. Based on the assumption of continuous fading, a multiple symbol differential automatic sphere decoding （MSDASD） algorithm is developed by incorporating a recursive form of an ML metric into automatic SD （ASD） algorithm. Furthermore, two algorithms, termed as MSD approximate ASD （MSDAASD） and MSD pruning ASD （MSDPASD）, are proposed to reduce computational complexity and the number of comparisons, respectively. Compared with the existing typical algorithms, i.e., multiple symbol differential feedback detection （MS-DFD） and noncoherent sequence detection （NSD）, the performance of the proposed algorithms is much superior to that of MS-DFD and a little inferior to that of NSD, while the complexity is lower than that of MS-DFD in most cases and significantly lower than that of NSD.
