基于NUMA架构的解释器访存优化设计与实现  

Design and implementation of memory optimization for NUMA based Interpreters

作　　者：任彤[1,2,3] 傅杰[1,2,4] 靳国杰 

机构地区：[1]中国科学院大学,北京100049 [2]中国科学院计算技术研究所,北京100190 [3]中国电子设备系统工程公司研究所,北京100141 [4]计算机系统结构国家重点实验室,北京100190 [5]龙芯中科技术有限公司,北京100190

出　　处：《高技术通讯》2015年第7期661-668,共8页Chinese High Technology Letters

基　　金：国家“核高基”科技重大专项课题(2009ZX01028-002-003,2009ZX01029-001-003,2010ZX01036-001-002,2012ZX01029-001-002-002,2014ZX01020201);国家自然科学基金(61221062,61133004,61173001,61232009,61222204,61432016);863计划(2012AA010901,2012AA011002,2013AA014301)资助项目

摘　　要：为了提高非一致内存访问(NUMA)架构虚拟机解释器的访存性能,研究了解释器在NUMA架构下的访存优化技术,提出了一种NUMA架构下的解释器访存优化方案,而且设计并实现了解释器的静态指令分派优化方法和动态指令分派优化方法。根据这一方案虚拟机启动时首先获取NUMA节点信息,并在每个NUMA节点中自动生成解释器所需的全部数据结构;解释器在运行时,通过静态或动态的指令分派技术来实现其执行线程在NUMA节点上访存的局部化。试验结果表明,上述方法能够显著提升解释器在NUMA系统中的性能。在DaCapo测试集上的总体性能提升了8%,最高性能提升幅度高达23%,而且算法实现代价低,适用于绝大多数的NUMA服务器系统。In order to improve the performance of a virtual machine＇ s interpreter under the non-uniform memory access （NUMA） architecture, a study of memory access optimization was conducted. Then, a scheme for memory optimi- zation of the interpreters under the NUMA architecture was proposed, and based on it, two novel approaches for an interpreter＇s memory optimization, namely the one using static instruction dispatching and another using dynamic instruction dispatching, were designed and implemented. According to the scheme, the virtual machine collects the information of the NUMA nodes when it starts up, and automatically generates all the data structures the interpreter needs in each NUMA node; when the interpreter is running on a NUMA node, it uses the two mentioned dispatc- hing optimization approaches to realize its local access to the NUMA nodes. The experimental results demonstrated that the proposed scheme can significantly improve the performance of interpreters in NUMA systems. The results of the experiment using the DaCapo showed that the overall performance of the interpreter was improved by 8 %, and the highest increment was up to 23%. The proposed memory optimization algorithm is very easyto implement and can be applied to most NUMA systems.

关 键 词：非一致内存访问(NUMA) 虚拟机 解释器 响应速度 启动性能 访存优化 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]