内存过载下基于重删的内存优化策略  

Memory Optimization Strategy Based on Deduplication Under Memory Overload

作　　者：刘瑞锴 邓玉辉[1,2] 

机构地区：[1]暨南大学信息科学技术学院,广州510632 [2]中国科学院计算技术研究所计算机体系结构国家重点实验室,北京100190

出　　处：《小型微型计算机系统》2017年第6期1243-1249,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61572232;61272073)资助;广东省自然科学基金重点项目(S2013020012865)资助;中国科学院计算机系统结构国家重点实验室开放基金项目(CARCH201401)资助

摘　　要：内存空间紧张下频繁的外存交换严重影响程序的运行性能,在内存过载时提高内存利用率和减少外存交换问题上,现有的研究主要集中在内存压缩和去重两个方面,压缩只能删除页内重复数据,不能做到页与页之间的相同数据删除;将去重技术与系统虚拟内存结合,删除换出页中的重复页,仍不能有效避免外存交换区访问.提出一种内存过载下基于重删的内存优化策略.利用页框回收算法选出适合去重的内存页,过滤其中全零页以减少存储和去重计算开销,删除重复页以提高内存利用率,并通过存储唯一数据页于内存去重区,有效降低外存交换操作,同时提供可配置的去重内存空间占用上限,避免过度去重引入的颠簸问题.在IO密集型、计算密集型、及常规应用场景下的实验测试表明,该策略能有效删除重复内存页,可以极大程度地减少外存交换操作,同时提高程序运行性能最高可达54.013%.Swapping page frequently when memory space is tight seriously affect the program performance, how to improve the memory utilization to reduce the external memory IO is a topic worthy of study. Previous studies mainly concentrated in memory page compression and deduplication,compression of memory page can only delete duplicate data within the page rather than the data between pages. Combined the deduplication technology with the virtual memory system, deleting the duplicate page that will be swap out, writing the remaining pages to swap area. However, the method still not an effective way to avoid external storage access. This paper presents a redundant memory page deleting mechanism under the memory overload scenario. Using the page frame recovery algorithm to choose the memory pages for deduplication, filtering the zero-page to reduce storage and calculation, removing duplicate pages in order to improve the memory utilization, and reducing the swap operation effectively by storing the remaining page in memory zone which is used for deduplication. Experimental tests under a variety of application scenarios show that the strategy can effectively remove duplicate memory pages, greatly reduce the swap operations, and improve program performance up to 54.013 %.

关 键 词：重复数据删除 内存交换操作 全零页过滤 哈希表 

分 类 号：TP333[自动化与计算机技术—计算机系统结构]