面向数据库内深度学习的基础计算体系及实现方案性能比较  

In-database Deep Learning Oriented Basic Computing System and Performance Comparison On Different Implementation Schemes

作　　者：樊宣伯 宋明阳 XUANBO Fan;SONG Mingyang(Harbin Institute of Technology,Harbin,Heilongjiang 150001,China)

机构地区：[1]哈尔滨工业大学,黑龙江哈尔滨150001

出　　处：《长江信息通信》2022年第10期77-79,共3页Changjiang Information & Communications

摘　　要：深度学习在海量数据分析中发挥着重要作用。随着数据量的快速增长,将训练数据从数据库移出会造成可观的时间损耗。尽管越来越多的数据库管理系统支持数据库内机器学习,但仍难以适应发展迅速的深度学习网络。因此,数据库管理系统有必要构建面向深度学习的基础运算体系,对外提供高性能的机器学习的基础算子,以支持更灵活的网络构建。文章讨论了基础体系的建立,对算子的执行性能建模,据此讨论了多种算子实现方案,测试了不同方案的性能。实验发现,基于内存整表存储的物理算子性能比其它方案有至少两倍的性能提升。Deep learning plays an essential role in massive data analysis.However,transferring data out of the database is timeconsuming.Although many DBMSs support in-database machine learning,it is not easy to adapt to the rapid development of deep learning.Therefore,DBMSs must build a basic computing architecture and provide high-performance operators for in-database deep learning.This paper discusses the basic computing architecture,presents a model of the operators’execution performance,discusses various operator implementation schemes,and tests the actual performance of different schemes.We found that the performance of physical operators based on the in-memory-whole-table storage scheme is at least two times better than other schemes.

关 键 词：智能数据库 深度学习 性能比较 基础运算 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]