Abstract:
With the continuous expansion of computer access to the sudden-centralized problems,traditional Least Recency Used(LRU),Least Frequency Used(LFU) and other cache replacement algorithms are difficult to meet the requirements of high hit ratio and low latency.According to the characteristics of the sudden-centralized data access mode,and the access mode influence to the data content popularity trend,this paper designs a new replacement strategy.The replacement strategy uses cache access count,access time,popularity prediction and others cache information to update the priority periodically,which is used to choose the replacement data.Meanwhile,according to the characteristics of the LRU,LFU and LIRS replacement algorithm,this paper compares the new strategy design idea,and the key factors of its replacement priority when the data contents are in different cases.Simulation results on the simulator SimpleScalar shows that the strategies is better than the traditional cache replacement strategy in sudden-centralized access mode.
Key words:
sudden-centralized access,
popularity prediction,
periodicity,
replacement priority,
hit ratio
摘要： 随着计算机系统对突发集中性问题访问规模的不断扩大,传统的最近最少使用(LRU)、最近最不常使用(LFU)等缓存替换算法已经难以满足高命中率、低延迟的要求。为此,针对数据突发集中性访问模式的特点,基于该模式对数据内容流行度变化趋势的影响,设计一种突发集中性访问模式的策略。该策略根据缓存的访问次数、访问时间、流行度预测等缓存信息,周期性地更新数据内容的置换优先级。同时对比LRU,LFU,LIRS及新策略在各种情况下,决定其数据内容置换优先级的因素。在模拟器SimpleScalar上的仿真结果表明,该策略在突发集中性访问模式下的性能优于传统的缓存替换策略。
关键词:
突发集中性访问,
流行度预测,
周期性,
置换优先级,
命中率
CLC Number: