面向大数据应用的众核处理器缓存结构设计

J4 ›› 2015, Vol. 37 ›› Issue (01): 28-35.

面向大数据应用的众核处理器缓存结构设计

万虎1,徐远超1,2,孙凤芸1,闫俊峰1

(1.首都师范大学信息工程学院,北京 100048;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)

收稿日期:2014-05-17 修回日期:2014-10-20 出版日期:2015-01-25 发布日期:2015-01-25
基金资助:
北京市自然科学基金资助项目（4143060）；北京市教委科技发展面上资助项目（KM201210028004）；计算机体系结构国家重点实验室开放课题（CARCH201203）；“高可靠嵌入式系统技术”北京市工程研究中心；北京市属高等学校人才强教项目国外访学（135300100）

Cache structure design for big data oriented many-core processor

WAN Hu1,XU Yuanchao1,2,SUN Fengyun1,YAN Junfeng1

(1.College of Information Engineering,Capital Normal University,Beijing 100048;2.State Key Laboratory of Computer Architecture,Institute of Computing Technology,
Chinese Academy of Sciences, Beijing 100190,China)

Received:2014-05-17 Revised:2014-10-20 Online:2015-01-25 Published:2015-01-25

摘要/Abstract

摘要：

大规模数据排序、搜索引擎、流媒体等大数据应用在面向延迟的多核/众核处理器上运行时资源利用率低下，一级缓存命中率高，二级/三级缓存命中率低，LLC容量的增加对IPC的提升并不明显。针对缓存资源利用率低的问题，分析了大数据应用的访存行为特点，提出了针对大数据应用的两种众核处理器缓存结构设计方案，两种结构均只有一级缓存，Share结构为完全共享缓存，Partition结构为部分共享缓存。评估结果表明，两种方案在访存延迟增加不多的前提下能大幅节省芯片面积，其中缓存容量较低时，Partition结构优于Share结构，缓存容量较高时，Share结构要逐渐优于Partition结构。由于众核处理器中分配到每个处理器核的容量有限，因此Partition结构有一定的优势。

关键词: 众核处理器, 大数据应用, 缓存设计, 访存行为, 数据中心

Abstract:

Some big data applications such as data sorting, search engine, streaming media running on the traditional latencyoriented multi/manycore processor are inefficiency. The hit rate of L1 Cache is high while that of L2/L3 Cache is relative low and IPC is not sensitive to LLC capacity. To address the low utilization issue of cache resources, we analyze the memory access patterns of big data applications, and then propose an optimization method of cache structure for manycore processor. Both the two structures only have L1 cache, while one is fully shared cache structure, and the other is partly shared cache partition structure. The evaluation results show that these two schemes can significantly save chip area at the cost of slightly increase of memory access. When cache capacity is low, the partition structure is superior to the share structure. As cache capacity increases, the share structure will gradually become superior to the partition structure. For manycore processors, the capacity assigned to each processor is limited, thus the partition structure has certain advantages.

Key words: many-core processor;big data application;cache design;memory access behavior;data center

万虎1,徐远超1,2,孙凤芸1,闫俊峰1. 面向大数据应用的众核处理器缓存结构设计[J]. J4, 2015, 37(01): 28-35.

WAN Hu1,XU Yuanchao1,2,SUN Fengyun1,YAN Junfeng1. Cache structure design for big data oriented many-core processor [J]. J4, 2015, 37(01): 28-35.

参考文献

［1］Wang Yuanzhuo,Jin Xiaolong,Cheng Xueqi.Network big data:Present and future ［J］. Chinese Journal of Computers, 2013, 36(6):115.(in Chinese)
［2］LotfiKamran P, Grot B, Ferdman M, et al. Scaleout processors［C］∥Proc of the 39th Annual International Symposium on Computer Architecture, 2012:500511.
［3］Ferdman M,Adileh A,Kocberber O,et al.Clearing the clouds:A study of emerging scaleout workloads on modern hardware ［C］∥Proc of the 17th International Conference on ASPLOS, 2012:3748.
［4］Chilimbi T M,Hill M D,Larus J R. Cacheconscious structure layout［J］.ACM SIGPLAN Notices, 1999, 34(5):112.
［5］Ding X,Wang K, Zhang X. ULCC:A userlevel facility for optimizing shared cache performance on multicores ［C］∥Proc of the 16th ACM Symposium on PPoPP, 2011:103112.
［6］Jia Y, Wu C, Zhang Z. Programs performance profiling optimization for guiding static cache partitioning［J］. Journal of Computer Research and Development, 2012,49(1):93102.(in Chinese)
［7］Kim S, Chandra D, Solihin Y. Fair cache sharing and partitioning in a chip multiprocessor architecture ［C］∥Proc of the 13th International Conference on PACT, 2004:111122.
［8］Tullsen D M, Eggers S J, Levy H M. Simultaneous multithreading:Maximizing onchip parallelism ［C］∥Proc of the 22nd Annual International Symposium on Computer Architecture,1995:533544.
［9］Zhan J,Zhang L,Sun N,et al.High volume throughput computing:Identifying and characterizing throughput oriented workloads in data centers［C］∥Proc of the IEEE 26th International Parallel and Distributed Processing Symposium Workshops & PhD Forum, 2012:17121721.
［10］Bienia C, Kumar S. The PARSEC benchmark suite:Characterization and architectural implications［C］∥Proc of the 17th ACM International Conference on Parallel Architectures and Compilation Techniques, 2008:122.
［11］Chen Y, Li W, Kim C, et al. Efficient shared Cache management through sharingaware replacement and streamingaware insertion policy ［C］∥Proc of IEEE International Parallel & Distributed Processing Symposium, 2009:111.
［12］Luk C K, Cohn R, Muth R, et al, Pin:Building customized program analysis tools with dynamic instrumentation ［C］∥Proc of the 2005 ACM SIGPLAN Conference on PLDI,2005:190200.
附中文参考文献：
［1］王元卓,靳小龙,程学旗. 网络大数据:现状与展望［J］.计算机学报, 2013,36(6):115.
［6］贾耀仓,武成岗,张兆庆. 指导Cache静态划分的程序性能profiling优化技术［J］. 计算机研究与发展,2012,49(1):93102.

[1]	陈冬林, 马一帆, 邹安琪, . 基于可再生能源消纳的数据中心负荷调度研究综述[J]. 计算机工程与科学, 2023, 45(05): 790-801.
[2]	刘忠沛, 吕高锋, 王继昌, 杨翔瑞. 专用数据处理器综述[J]. 计算机工程与科学, 2023, 45(02): 215-227.
[3]	曹继军. 面向HPC和DC的可重构光互连网络体系结构综述[J]. 计算机工程与科学, 2022, 44(06): 951-963.
[4]	段晨, 彭伟, 王宝生. 数据中心网络路由研究进展[J]. 计算机工程与科学, 2022, 44(04): 631-644.
[5]	李力, 汪硕, 黄韬, 刘韵洁, . 数据中心网络四层负载均衡技术综述[J]. 计算机工程与科学, 2022, 44(01): 48-59.
[6]	毛媛媛, 石恩雅, 蒋从锋, 仇烨亮, 贾刚勇, 万健, 闫龙川. 基于遗传算法的数据中心能效仿真[J]. 计算机工程与科学, 2021, 43(08): 1341-1352.
[7]	路煜, 张路, 侯小凤, 郑文立, 李超. 数据中心功耗削峰电池的可用性分析[J]. 计算机工程与科学, 2021, 43(01): 9-16.
[8]	韩彪, 王韬, 王宝生. 无线数据中心网络：进展、挑战和展望[J]. 计算机工程与科学, 2021, 43(01): 70-81.
[9]	梁毅，曾绍康，梁岩德，丁毅. 一种基于周期性特征的数据中心在线负载资源预测方法[J]. 计算机工程与科学, 2020, 42(03): 381-390.
[10]	王济伟1，葛浙奉1，蒋从锋1，张纪林1，俞俊1，林江彬2，闫龙川3，任祖杰4，万健5. 混部数据中心负载特征及其任务调度优化分析[J]. 计算机工程与科学, 2020, 42(01): 8-17.
[11]	刘少楠，李玲，苑迎，蒋国佳，王聪，吕艳霞. 基于节点连通性排序的虚拟网络映射算法[J]. 计算机工程与科学, 2019, 41(12): 2101-2109.
[12]	王武1，王舒扬1,2，姜金荣1,孟虹松3. 快速多极子方法在申威众核处理器上的实现和优化[J]. 计算机工程与科学, 2019, 41(07): 1161-1167.
[13]	陈逸飞，朱蕾，李宏亮. 一种多线程阵列众核处理器的二级Cache划分机制[J]. 计算机工程与科学, 2019, 41(03): 400-408.
[14]	李树楠1,2,3，詹男杰1,2,3，章玥1,2. 面向数据中心租户带宽特征的虚拟拥塞控制方法性能分析[J]. 计算机工程与科学, 2018, 40(09): 1562-1571.
[15]	樊自甫，张丹，李书. 基于软件定义网络的数据中心网络负载均衡算法研究[J]. 计算机工程与科学, 2018, 40(06): 1017-1022.