Daniel Schall

Energy efficiency is not enough, energy proportionality is needed!

Due to the energy consumption/resource utilization characteristics of todays centralized DB servers, the fastest configuration is also the most energy-efficient one. Extensive use of SSDs alone cannot enable a fundamental change of this overall picture, because the storagerelated energy consumption is typically only a little fraction of the overall energy budget. Even, when this storage-related share is (almost) completely reduced by optimized flash-aware buffer management, the saving effect achieved may be limited by less than ∼10%. Therefore, we have designed a cluster of wimpy computing nodes called WattDB, where the individual nodes are dynamically attached and detached to the cluster on demand - depending on the current workload needs -, thereby aiming at energy-proportional DB management.

Enhancing energy efficiency of database applications using SSDs

Presently, solid state disks (SSDs) are emerging as a disruptive storage technology and promise breakthroughs for important application properties. They quickly enter the enterprise domain and (partially) replace magnetic disks (HDDs) for database servers. To identify performance and energy use of both types of storage devices, we have built an analysis tool and measured access times and energy needed for them. Associating these measurements to physical IO patterns, we checked and verified the performance claims given by the device manufacturers. Using typical read/write access patterns frequently observed in IO-intensive database applications, we fathomed the performance and energy efficiency potential of a spectrum of differing storage devices (low-end, medium, and high-end SSDs and HDDs). Cross-comparing measurements of identical experiments, we present indicative parameters concerning IO performance and energy consumption. Furthermore, we reexamine an IO rule of thumb guiding their energy-efficient use in database servers. These findings suggest some database-related optimization areas where they can improve performance while energy is saved at the same time.

Towards an enhanced benchmark advocating energy-efficient systems

The growing energy consumption of data centers has become an area of research interest lately. For this reason, the research focus has broadened from a solely performance-oriented system evaluation to an exploration where energy efficiency is considered as well. The Transaction Processing Performance Council (TPC) has also reflected this shift by introducing the TPC-Energy benchmark. In this paper, we recommend extensions, refinements, and variations for such benchmarks. For this purpose, we present performance measurements of real-world DB servers and show that their mean utilization is far from peak and, thus, benchmarking results, even in conjunction with TPC-Energy, lead to inadequate assessment decisions, e.g., when a database server has to be purchased. Therefore, we propose a new kind of benchmarking paradigm that includes more realistic power measures. Our proposal will enable appraisals of database servers based on broader requirement profiles instead of focusing on sole performance. Furthermore, our energy-centric benchmarks will encourage the design and development of energy-proportional hardware and the evolution of energy-aware DBMSs.

WattDB: an energy-proportional cluster of wimpy nodes

The constant growth of data in all businesses leads to bigger database servers. While peak load times require fast and heavyweight hardware to guarantee performance, idle times are a waste of energy and money. Todays DBMSs have the ability to cluster several servers for performance and fault tolerance. Nevertheless, they do not support dynamic powering of the clusters nodes based on the current workload. In this demo, we propose a newly developed DBMS running on clustered commodity hardware, which is able to dynamically power nodes. The demo allows the user to interact with the DBMS and adjust workloads, while the clusters reaction is shown in real-time.

Performance and power evaluation of flash-aware buffer algorithms

With flash disks being an important alternative to conventional magnetic disks, various design aspects of DBMSs, whose I/O behavior is performance-critical, and especially their I/O architecture should be reconsidered. Taking the distinguished characteristics of flash disks into account, several flash-aware buffer algorithms have been proposed with focus on flash-specific performance optimizations. We introduce several basic principles of flash-aware buffer management and evaluate performance and energy consumption of related algorithms in a DBMS environment using both flash disks and magnetic disks. Our experiments reveal the importance of those principles and the potential of flash disks both in performance improvement and in energy saving.

Approximating an Energy-Proportional DBMS by a Dynamic Cluster of Nodes

The most energy-efficient configuration of a single-server DBMS is the highest performing one, if we exclusively focus on specific applications where the DBMS can steadily run in the peak-performance range. However, typical DBMS activity levels—or their average system utilization—are much lower and their energy use is far from being energy proportional. Built of commodity hardware, WattDB—a distributed DBMS—runs on a cluster of computing nodes where energy proportionality is approached by dynamically adapting the cluster size. In this work, we combine our previous findings on energy-proportional storage layers and query processing into a single, transactional DBMS. We verify our vision by a series of benchmarks running OLTP and OLAP queries with varying %intensity. degrees of parallelism. These experiments illustrate that WattDB dynamically adjusts to the workload present and reconfigures itself to satisfy performance demands while keeping its energy consumption at a minimum.

WattDB - A Journey towards Energy Efficiency

Due to their narrow power spectrum between idle and full utilization [2], satisfactory energy efficiency of servers can only be reached in the peak-performance range, whereas energy efficiency obtained for lower activity levels is far from being optimal. Hence, this hardware property obviates a desired energy proportionality or minimal energy use for the entire range of system utilization. To approximate energy proportionality for all activity levels, we developed various versions of WattDB, a distributed DBMS, which runs on a dynamic cluster of wimpy computing nodes. In this survey, we sketch important design decisions and implementation steps towards the final state of WattDB. For these reasons, we discuss our findings on a cluster with dedicated storage nodes and static data allocation, on dynamic data repartitioning and allocation, and on a dynamic cluster where each node can serve as storage and processing node in a symmetric way. Our experiments show that WattDB dynamically adjusts to the workload present and reconfigures itself to satisfy performance demands while keeping its energy consumption at a minimum. Finally, we compare the performance and energy results of the WattDB software running on the cluster of wimpy nodes with that of a brawny server.

The SIGMOD 2010 programming contest a distributed query engine

We report on the second annual ACM SIGMOD programming contest, which consisted in building an efficient distributed query engine on top of an in-memory index. This article is co-authored by the organizers of the competition (Clément Genzmer, Pierre Senellart) and the students who built the two leading implementations (Volker Hudlet, Hyunjung Park, Daniel Schall).

„Von der Torfabrik zur Denkfabrik“ Bericht zur 14. Fachtagung „Datenbanksysteme für Business, Technologie und Web“

Seit 1985 richtet der GI-Fachbereich Datenbanken und Informationssysteme (DBIS) alle zwei Jahre seine Fachtagung Datenbanksysteme fur Business, Technologie und Web (BTW) aus, die in diesem Jahr vom 28. Februar bis zum 4. Marz in Kaiserslautern durchgefuhrt wurde. Nach 1991, als die 4. Fachtagung dieser Reihe an der TU Kaiserslautern stattfand, durfte die Arbeitsgruppe DBIS des Fachbereichs Informatik der TU damit zum zweiten Mal die groste Wissenschaftsveranstaltung der deutschsprachigen Datenbankgemeinde ausrichten. Sie wurde von Prof. Dr-.Ing. Dr. h.c. Theo Harder (lokale Organisation) und zwei „Ehemaligen“ der TU Kaiserslautern, Prof. Dr.-Ing. habil. Bernhard Mitschang (Wissenschaftliches Programmkomitee, Universitat Stuttgart) und Dr.-Ing. Harald Schoning (IndustrieProgrammkomitee, Software AG, Darmstadt), organisiert. Jurgen Bittner, Vorstandsvorsitzender der SQL Projekt AG, Dresden, und Prof. Dr. Hans-Jurgen Schek, Emeritus der ETH Zurich, waren als Ehrengaste der BTW eingeladen. Beide hatten 1991 zur 4. BTW-Tagung die Hauptvortrage mit den Themen „Die Architekturkonzeption eines DBMS aus pragmatischer Sicht“ und „Erweiterbarkeit, Kooperation, Foderation von Datenbanksystemen“ gehalten. Das Fachprogramm bestand aus mehr als 40 Vortragen im Wissenschaftsund Industrieprogramm und 12 Demonstrationen von neuen Informatik-Anwendungen. Alle Beitrage wurden in einem anonymen Begutachtungsverfahren von uber 50 Fachleuten aus dem DBIS-Bereich ausgewahlt. Die