Online Updates on Data Warehouses via Judicious Use.pdf
免费下载
6
Online Updates on Data Warehouses via Judicious Use
of Solid-State Storage
MANOS ATHANASSOULIS,
´
Ecole Polytechnique F
´
ed
´
eraledeLausanne
SHIMIN CHEN, Chinese Academy of Sciences
ANASTASIA AILAMAKI,
´
Ecole Polytechnique F
´
ed
´
eraledeLausanne
PHILIP B. GIBBONS, Intel Labs, Pittsburgh
RADU STOICA,
´
Ecole Polytechnique F
´
ed
´
eraledeLausanne
Data warehouses have been traditionally optimized for read-only query performance, allowing only offline
updates at night, essentially trading off data freshness for performance. The need for 24x7 operations in
global markets and the rise of online and other quickly reacting businesses make concurrent online up-
dates increasingly desirable. Unfortunately, state-of-the-art approaches fall short of supporting fast analysis
queries over fresh data. The conventional approach of performing updates in place can dramatically slow
down query performance, while prior proposals using differential updates either require large in-memory
buffers or may incur significant update migration cost.
This article presents a novel approach for supporting online updates in data warehouses that overcomes
the limitations of prior approaches by making judicious use of available SSDs to cache incoming updates.
We model the problem of query processing with differential updates as a type of outer join between the data
residing on disks and the updates residing on SSDs. We present MaSM algorithms for performing such joins
and periodic migrations, with small memory footprints, low query overhead, low SSD writes, efficient in-place
migration of updates, and correct ACID support. We present detailed modeling of the proposed approach,
and provide proofs regarding the fundamental properties of the MaSM algorithms. Our experimentation
shows that MaSM incurs only up to 7% overhead both on synthetic range scans (varying range size from
4KB to 100GB) and in a TPC-H query replay study, while also increasing the update throughput by orders
of magnitude.
Categories and Subject Descriptors: H.2.4 [Database Management]: Systems—Query processing; H.2.7
[Database Management]: Database Administration—Data warehouse and repository
General Terms: Algorithms, Design, Performance
Additional Key Words and Phrases: Materialized sort merge, online updates, data warehouses, SSD
An earlier version of this article appeared in the 2011 Proceedings of the ACM SIGMOD International
Conference on Management of Data.
M. Athanassoulis is currently affiliated with Harvard University. R. Stoica is currently affiliated with IBM
Research, Zurich.
This work was partially supported by an ESF EurYI award (FN 511.261), NSF funds, an EU-funded FP7
project (grant no. 317858), and the Intel Science & Technology Center for Cloud Computing. S. Chen is
supported by the CAS Hundred Talents program and by NSFC Innovation Research Group no. 61221062.
Authors’ addresses: M. Athanassoulis (corresponding author), School of Engineering and Applied Sciences,
Harvard University, Cambridge, MA 02138; email: manos@seas.harvard.edu; S. Chen (corresponding au-
thor), State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy
of Sciences, no. 6 Kexueyuan South Road, Zhongguancun, Haidian District Beijing, China; A. Ailamaki,
School of Computer and Communication Sciences,
´
Ecole Polytechnique F
´
ed
´
erale de Lausanne, Lausanne,
Switzerland; P. B. Gibbons, Intel Labs Pittsburgh and Carnegie Mellon University, 5000 Forbes Avenue,
Pittsburgh, PA 15213; R. Stoica, IBM Research Zurich, Saumerstrasse 4, 8803 Ruschlikon, Switzerland.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by
others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions
from Permissions@acm.org.
c
2015 ACM 0362-5915/2015/03-ART6 $15.00
DOI: http://dx.doi.org/10.1145/2699484
ACM Transactions on Database Systems, Vol. 40, No. 1, Article 6, Publication date: March 2015.
6:2 M. Athanassoulis et al.
ACM Reference Format:
Manos Athanassoulis, Shimin Chen, Anastasia Ailamaki, Philip B. Gibbons, and Radu Stoica. 2015. Online
updates on data warehouses via judicious use of solid-state storage. ACM Trans. Datab. Syst. 40, 1, Article 6
(March 2015), 42 pages.
DOI: http://dx.doi.org/10.1145/2699484
1. INTRODUCTION
Data warehouses (DWs) are typically designed for efficient processing of read-only anal-
ysis queries over large data. Historically, updates to the data were performed using
bulk insert/update features that executed offline—mainly during extensive idle times
(e.g., at night). Two important trends lead to a need for a tighter interleaving of anal-
ysis queries and updates. First, the globalization of business enterprises means that
analysis queries are executed round-the-clock, eliminating any idle-time window that
could be dedicated to updates. Second, the rise of online and other quickly reacting
businesses means that it is no longer acceptable to delay updates for hours as older
systems did: the business value of the answer often drops precipitously as the under-
lying data becomes more out of date [Inmon et al. 2003; White 2002]. In response to
these trends, data warehouses must now support a much tighter interleaving of anal-
ysis queries and updates, so that analysis queries can occur 24/7 and take into account
very recent data updates [Becla and Lim 2008]. The large influx of data is recognized
as one of the key characteristics of modern workloads, often referred to as velocity of
data [Zikopoulos et al. 2012]. Thus, active (or real-time) data warehousing has emerged
as both a research topic [Polyzotis et al. 2008; Athanassoulis et al. 2011] and a busi-
ness objective [Oracle 2013; White 2002; IBM 2013; Russom 2012] aiming to meet the
increasing demands of applications for the latest version of data. Unfortunately, state-
of-the-art data warehouse management systems fall short of the business goal of fast
analysis queries over fresh data. A key unsolved problem is how to efficiently execute
analysis queries in the presence of online updates that are needed to preserve data
freshness.
1.1. Efficient Online Updates for DW: Limitations of Prior Approaches
While updates can proceed concurrently with analysis queries using concurrency con-
trol schemes such as snapshot isolation [Berenson et al. 1995], the main limiting factor
is the physical interference between concurrent queries and updates. We consider the
two known approaches for supporting online updates—in-place updates and differen-
tial updates—and discuss their limitations.
In-Place Updates Double Query Time. A traditional approach used in OLTP systems
is to update in place, that is, to store the new value in the same physical location as the
previous one. However, as shown in Section 2.2, in-place updates can dramatically slow
down data warehousing queries. Mixing random in-place updates with TPC-H queries
increases the execution time, on average, by 2.2x on a commercial row-store data
warehouse and by 2.6x on a commercial column-store data warehouse. In the worst
case, the execution time is 4x longer. Besides having to service a second workload
(i.e., the updates), the I/O subsystem suffers from the interference between the two
workloads: the disk-friendly sequential scan patterns of the queries are disrupted by
the online random updates. This factor alone accounts for 1.6x slowdown on average
in the row-store DW.
Differential Updates Limited by In-Memory Buffer or HDD Performance. Recently,
differential updates have been proposed as a means to enable efficient online updates
in column-store data warehouses [H
´
eman et al. 2010; Stonebraker et al. 2005], follow-
ing the principle of differential files [Severance and Lohman 1976]. The basic idea is
to: (i) cache incoming updates in an in-memory buffer; (ii) take the cached updates
ACM Transactions on Database Systems, Vol. 40, No. 1, Article 6, Publication date: March 2015.
of 42
免费下载
【版权声明】本文为墨天轮用户原创内容，转载时必须标注文档的来源（墨天轮），文档链接，文档作者等基本信息，否则作者和墨天轮有权追究责任。如果您发现墨天轮中有涉嫌抄袭或者侵权的内容，欢迎发送邮件至：contact@modb.pro进行举报，并提供相关证据，一经查实，墨天轮将立刻删除相关内容。
下载排行榜
评论