利用堆棧特征的片上末級緩存訪問模式在線識別方法
發(fā)布時間:2018-09-12 12:21
【摘要】:很多優(yōu)化處理器緩存利用效率的方法依賴于對訪問請求序列的特征的探測或識別,例如,預取和繞開等。如何在線有效識別訪問請求序列的特征依然是一個開放的問題。通過對典型訪問模式的深入分析,發(fā)現(xiàn)其堆棧距離頻度的分布展示出鮮明的特征。而模擬實驗數(shù)據(jù)表明訪問請求序列的特征具有一定的持續(xù)性和穩(wěn)定性,具有檢測和預測的可行性。因而提出了一種基于堆棧直方圖峰值的在線識別訪問模式的機制和方法,空間和時間開銷都較小。對SPEC CPU2000/2006的15個程序的實驗表明,所提方法均可正確識別測試程序的訪問模式。
[Abstract]:Many methods to optimize the efficiency of processor cache utilization depend on the detection or recognition of the features of access request sequences, such as prefetching and bypassing. How to effectively identify the features of access request sequences is still an open problem. Through in-depth analysis of typical access patterns, it is found that the distribution of stack distance frequency shows distinct characteristics. The simulation results show that the characteristics of access request sequences are consistent and stable, and have the feasibility of detection and prediction. Therefore, a mechanism and method for on-line identification of access patterns based on stack histogram peak is proposed, with less space and time overhead. Experiments on 15 programs of SPEC CPU2000/2006 show that the proposed method can correctly identify the access patterns of test programs.
【作者單位】: 北京郵電大學計算機學院;北京航空航天大學軟件開發(fā)環(huán)境國家重點實驗室;吉林師范大學數(shù)學學院;
【基金】:中國博士后基金資助項目(2014M550662) 軟件開發(fā)環(huán)境國家重點實驗室資助項目(SKLSDE-2014KF-04)
【分類號】:TP332
本文編號:2238989
[Abstract]:Many methods to optimize the efficiency of processor cache utilization depend on the detection or recognition of the features of access request sequences, such as prefetching and bypassing. How to effectively identify the features of access request sequences is still an open problem. Through in-depth analysis of typical access patterns, it is found that the distribution of stack distance frequency shows distinct characteristics. The simulation results show that the characteristics of access request sequences are consistent and stable, and have the feasibility of detection and prediction. Therefore, a mechanism and method for on-line identification of access patterns based on stack histogram peak is proposed, with less space and time overhead. Experiments on 15 programs of SPEC CPU2000/2006 show that the proposed method can correctly identify the access patterns of test programs.
【作者單位】: 北京郵電大學計算機學院;北京航空航天大學軟件開發(fā)環(huán)境國家重點實驗室;吉林師范大學數(shù)學學院;
【基金】:中國博士后基金資助項目(2014M550662) 軟件開發(fā)環(huán)境國家重點實驗室資助項目(SKLSDE-2014KF-04)
【分類號】:TP332
【相似文獻】
相關期刊論文 前3條
1 李紅衛(wèi);黃卓恒;李翠萍;;云存儲中數(shù)據(jù)安全訪問的研究[J];江蘇技術師范學院學報;2013年04期
2 李紅衛(wèi);古春生;景征駿;于志敏;;云存儲中基于ORAM的數(shù)據(jù)安全訪問[J];微電子學與計算機;2014年06期
3 ;[J];;年期
相關會議論文 前1條
1 李宇騫;袁欣顥;王磊;劉超;劉洋;顧榮輝;王乃錚;陳渝;;分布式共享設備的等時傳輸優(yōu)化[A];第六屆和諧人機環(huán)境聯(lián)合學術會議(HHME2010)、第19屆全國多媒體學術會議(NCMT2010)、第6屆全國人機交互學術會議(CHCI2010)、第5屆全國普適計算學術會議(PCC2010)論文集[C];2010年
,本文編號:2238989
本文鏈接:http://sikaile.net/kejilunwen/jisuanjikexuelunwen/2238989.html
最近更新
教材專著