摘要
Skein算法提供一种树形哈希模式,方便了并行硬件资源的充分利用.提出一种并行计算平台,用以评估树形哈希算法与传统线性哈希算法相比的性能提升.该平台通过开关网络连接多个运算单元和多个存储单元,允许并行运算和并发的存储器访问.平台中包含一个控制器,负责动态地向运算单元调度任务,支持算法参数的灵活配置.为了确定最优配置参数,根据平台特征建立了性能模型.实验结果显示,加速比趋近于理论上限,即平台中运算单元的数量.该平台的硬件原型通过台积电65 nm工艺进行综合,工作频率达到833 MHz,对应吞吐率为38.091 Gbps.
Skein algorithm provides a tree hashing mode to make full use of parallel hardware resources. A platform with parallel architecture is presented to evaluate the performance improvement of tree hashing over normal serial hashing. This platform connects multiple execution units and memory units using a switch network, enabling concurrent computations and memory accesses. A controller in the platform performs the task scheduling for execution units dynamically, supporting the flexible parameter selection in tree hashing. To determine the optimal parameter selection, a performance model has been developed aiming at this platform. Experiments reveal that the speedup ratio achieved could approach to the theoretical upper bound with respect to the number of execution units. The hardware prototype of the proposed platform is synthesized in TSMC 65 nm CMOS process. It achieves a throughput of 38. 091 Gbps at the working frequency of 833 MHz.
出处
《复旦学报(自然科学版)》
CAS
CSCD
北大核心
2014年第1期37-50,共14页
Journal of Fudan University:Natural Science
基金
Project supported by the State Key Program of National Natural Science of China(61234002)