Datenbestand vom 20. Mai 2019

Warenkorb Datenschutzhinweis Dissertationsdruck Dissertationsverlag Institutsreihen     Preisrechner

aktualisiert am 20. Mai 2019

ISBN 9783843920858

Euro 84,00 inkl. 7% MwSt


978-3-8439-2085-8, Reihe Informatik

Daniel Schall
Energy Efficiency in Database Systems

247 Seiten, Dissertation Technische Universität Kaiserslautern (2015), Softcover, A5

Zusammenfassung / Abstract

Today, database management systems are highly complex, ubiquitous pieces of software. Traditionally, the main focus in DBMS development was perfomance-centric, hence, heavy-weight systems with peak performance were developed.

In the last years, electricity prices have increased significantly, and hence, the motivation to save power came into focus of datacenters and into the DBMS community. Recent ventures on increasing energy efficiency of databases did contribute only marginal improvements.

In this thesis, we present our findings on an elastic cluster of lightweight nodes, that is able to dynamically adapt to the workload by scaling out and back in. By powering down unused nodes, we avoid high idle power consumption of today’s hardware. Our contribution can be summarized as follows:

• We present a newly developed benchmarking paradigm for evaluating energy-efficiency, since existing DB benchmarks do not include the whole power spectrum in their measurements.

• We describe our work on measuring and benchmarking energy efficiency of commodity hardware with a custom-made measurement track, capable of metering power and energy consumption of an entire server cluster and correlating the readings with benchmark results.

• We studied and improved our theories on our own database management system, called WattDB, which we also present in detail in this thesis. WattDB is a distributed DBMS, running on lightweight, commodity hardware, with energy efficiency as prime optimization goal. Since the database is distributed among all nodes in the cluster, we also introduce techniques necessary for repartitioning to keep data available at all times. Finally, we compare our approach to a traditional server in terms of power and performance.