不对称内存计算平台OLAP查询处理技术研究

张延松; 张宇; 周烜; 王珊

doi:10.3969/j.issn.1000-5641.2016.05.011

不对称内存计算平台OLAP查询处理技术研究

doi: 10.3969/j.issn.1000-5641.2016.05.011

张延松^1,2,3,
张宇⁴,
周烜^1,2,
王珊^1,2

1.
中国人民大学 DEKE实验室, 北京 100872;
2.
中国人民大学信息学院, 北京 100872;
3.
中国人民大学中国调查与数据中心, 北京 100872;
4.
国家卫星气象中心, 北京 100081

基金项目:

国家 863 计划项目(2015AA015307);中央高校基本科研业务费专项资金项目(16XNLQ02);华为创新研究计划(HIRP 20140507, HIRP 20140510)

计量
- 文章访问数: 284
- HTML全文浏览量: 7
- PDF下载量: 527
- 被引次数: 0
出版历程
- 收稿日期: 2016-06-27
- 刊出日期: 2016-09-25

Research on OLAP query processing technology for asymmetric in-memory computing platform

1.
DEKE Lab, Renmin University of China, Beijing 100872, China;
2.
School of Information, Renmin University of China, Beijing 100872, China;
3.
National Survey Research Center at Renmin University of China, Beijing 100872, China;4. National Satellite Meteorological Center, Beijing 100081, China

摘要

摘要: 给出了一种面向当前和未来不对称内存计算平台的 OLAP 查询处理技术. 不对称内存计算平台是指配置有不同计算类型的处理器、不同存储访问设备的计算机, 因此需要对 OLAP 查询处理模型按不同的计算特点进行优化存储配置和实现算法设计, 从而使 OLAP 查询处理的不同阶段更好地适应相应的存储与计算设备的硬件特点, 提高硬件设备的利用率, 更好地发挥硬件的性能. 提出了 3 阶段 OLAP 计算模型, 将传统基于迭代处理模型的 OLAP 查询处理过程分解为计算密集型和数据密集型负载, 分别由功能完备的通用处理器和并行计算能力强大的协处理器分而治之地完成, 并最小化不同存储与计算设备之间的数据传输代价. 实验结果表明基于负载划分的 3 阶段 OLAP 计算模型能够较好地适应 CPU-Phi 不对称计算平台, 实现通过计算型硬件加速计算密集型负载, 从而加速整个 OLAP 查询处理性能的目标.
- 内存计算 /
- 不对称计算平台 /
- 内存联机分析处理
Abstract: This paper proposes an OLAP query processing technology for nowadays and future asymmetric in-memory computing platform. Asymmetric in-memorycomputing platform means that computer equips with different computing feature processors and different memory access devices so that the OLAP processing model needs to be optimized fordifferent computing features and implementation designs to enable the different processing stages to adapt to the characteristics of corresponding storage and computing hardware for higher hardware utilization and performance. This paper proposes the 3-stage OLAP computing model, which divides the traditional iterative processing model into computing intensive and data intensive workloads to be assigned to general purpose processor with full fledged functions and coprocessor with powerful parallel processing capacity. The data transmission overhead between different storage and computing devices is also minimized. The experimental results show that the 3-stage OLAP computing model based on workload partitioning can be adaptive to CPU-Phi asymmetric computing platform, the acceleration on OLAP query processing can be achieved by accelerating computing intensive workload by computing intensive hardware.
- in-memory computing asymmetric computing platform in-memory OLAP /

HTML全文

参考文献(1)

[1]

[ 1 ] SEBASTIAN ANTHONYIntel unveils new Xeon chip with integrated FPGA, touts 20x performance boost [EB/OL]. (2014-01-19)[2015-12-25]. http://www.extremetech.com/extreme/184828-intel-unveils-new-xeon-chip-with-integrated-fpga-touts-20x-performance-boost.
[ 2 ] JIM H. IBM launches flashDIMMs [EB/OL]. (2014-01-20)[2015-12-25]. http://thessdguy.com/ibm-launches-flash-dimms/.
[ 3 ] ANTON S. Intel: First 3D XPoint SSDs will feature up to 6GB/s of bandwidth [EB/OL]. (2015-08-28)[2016-03-16]. http://www.kitguru.net/components/memory/anton-shilov/intel-first-3d-xpoint-ssds-will-feature-up-to-6gbs-of-bandwidth/.
[ 4 ] BLANAS S, LI Y, PATEL J M. Design and evaluation of main memory hash join algorithms for multi-core CPUs [C]//SIGMOD. 2011: 37-48.
[ 5 ] BALKESEN C, TEUBNER J, ALONSO Get al Main-memory hash joins on multi-core cpus: Tuning to the underlying hardware [C]//ICDE. 2013: 362-373.
[ 6 ] ALBUTIU M-C, KEMPER A, NEUMANN T Massively parallel sort-merge joins in main memory multi-core data-base systems [J]. VLDB Endowment, 2012, 5(10): 1064-1075.
[ 7 ] HE B, YANG K, FANG Ret al. Relational joins on graphics processors [C]//SIGMOD. 2008: 511-524.
[ 8 ] YUAN Y, LEE R, ZHANG XThe yin and yang of processing data warehousing queries on GPU devices [J]. PVLDB, 2013, 6(10): 817-828.
[ 9 ] PIRK H, MANEGOLD S, KERSTEN M L. Accelerating foreign-key joins using asymmetric memory channels [C]//ADMS@VLDB. 2011: 27-35.
[10] HE J, LU M, HE B. Revisiting co-processing for hash joins on the coupled CPU-GPU architecture [J]. VLDB Endowment, 2013, 6(10): 889-900.
[11] JHA S, HE B, LU M, et al. Improving main memory hash joins on Intel Xeon Phi processors: an experimental approach [J]. Proceedings of TheVldb Endowment, 2015, 8(6): 642-653.
[12] POLYCHRONIOU O, RAGHAVAN A, ROSS K A. Rethinking SIMD vectorization for in-memory databases [C]//SIGMOD Conference. 2015: 1493-1508.
[13] HALSTEAD R J, ABSALYAMOV I, NAJJAR W A, et al. FPGA-based Multithreading for In-Memory Hash Joins [C]//Conference on Innovative Data Systems Research. 2015.
[14] ZHANG Y, ZHOU X, ZHANG Y, et al. Virtual Denormalization via Array Index Reference for Main Memory OLAP [J]. IEEE Transactions on Knowledge and Data Engineering, 2016, 28(4): 1061-1074.
[15] ALEKSIC S, CELIKOVIC M, LINK S, et al. Face off: Surrogate vs. natural keys [C]//Advances in Databases and Information Systems-14th East European Conference. 2010: 543-546.
[16] 张宇, 张延松, 陈红, 等. GPU semi-MOLAP：一种适应 GPU 的混合 OLAP 查询处理模型[J]. 软件学报,2016, 27(5): 1246-1265.