StreamDB: A Unified Data Management System for Service-Based Cloud Application
StreamDB: A Unified Data Management System for Service-Based Cloud Application
复制标题
DOI：
10.1109/scc.2018.00029
复制
发表时间：
2018-07
期刊：
影响因子：
--
通讯作者：
Huankai Chen;Matteo Migliavacca
中科院分区：
文献类型：
--
作者：
Huankai Chen;Matteo Migliavacca
目前的数据管理系统主要分为两大类：数据库管理系统（DBMS）和数据流管理系统（DSMS）。在现代基于服务的云应用程序中越来越多地使用流分析，这在DBMS供应商之间引发了一场军备竞赛，以提供更复杂的数据库内流支持，这需要处理快速数据收集的数量，种类，速度和可变性。不幸的是，当前的解决方案要么只提供有限的流分析能力和水平可扩展性（经典的RDBMS），要么为其他属性（NoSQL DBMS）权衡事务处理，导致DBMS没有“一刀切”的诅咒。在本文中，我们认为，交易处理是一个相关的概念DSMS。作为迈向“一刀切”数据管理系统的第一步，我们提出了StreamDB，它集成了DSMS中的事务处理，而不是扩展DBMS来支持流。首先，我们描述了StreamDB如何在流环境中处理事务，然后在典型的事务基准测试中将我们的方法与传统的内存中DBMS进行比较。我们的研究结果表明，StreamDB在吞吐量，可扩展性和延迟方面具有优势。最后，我们认为，这里提出的想法提供了深入了解下一代数据管理系统的发展，并激励进一步研究固有的挑战，统一DBMS和DSMS。
Current data management systems are mainly divided into two categories: Database Management System (DBMS) and Data Stream Management System (DSMS). The increasing use of streaming analysis in modern service-based cloud applications has created an arms race among DBMS vendors to offer ever more sophisticated in-database streaming support, which requires handling the volume, variety, velocity and variability of fast data collections. Unfortunately, current solutions either only provide limited streaming analysis capacity and horizontal scalability (classic RDBMS) or trade off transaction processing for other properties (NoSQL DBMS), leading to the curse of no "one size fits all" for DBMS. In this paper, we argue that transaction processing is a relevant concept for DSMS. As a first step toward "One Size Fits All" Data Management System, we present StreamDB, which integrates transaction processing in DSMS as opposed to extending DBMS to support streams. First, we describe how StreamDB processes transactions in a streaming environment, then we compare our approach with traditional in-memory DBMS on typical transactional benchmarks. Our results show that StreamDB is advantageous in terms of throughput, scalability, and latency. Finally, we argue that the ideas present here provide insight on the development of next-generation data management systems and motivate further study of the challenges inherent in unifying DBMS and DSMS.