Generic Online Optimization of Multiple Configuration Parameters with Application to a Database Server

Diao, Yixin; Eskesen, Frank; Froehlich, Steven; Hellerstein, Joseph L.; Spainhower, Lisa; Surendra, Maheswaran

doi:10.1007/978-3-540-39671-0_2

Cited by 23 publications

(12 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Our main idea in this article is that machine learning techniques can be used to extract interesting information about an algorithm's empirical hardness automatically from huge bodies of experimental data. (Our work can be seen as part of a broader trend that leverages machine learning in more applied areas of computer science (e.g., Diao et al [2003], Zheng et al [2003], Goldszmidt [2007], and especially Goldsmith et al [2007]). A particularly related early article is work by Brewer [1994Brewer [ , 1995 on using statistical models to automatically optimize high-level libraries for parallel processors; we describe it in detail in Section 6.2.1.1.)…”

Section: :3mentioning

confidence: 99%

Empirical hardness models

Leyton‐Brown

Nudelman

Shoham

2009

J. ACM

View full text Add to dashboard Cite

Is it possible to predict how long an algorithm will take to solve a previously-unseen instance of an NP-complete problem? If so, what uses can be found for models that make such predictions? This article provides answers to these questions and evaluates the answers experimentally.We propose the use of supervised machine learning to build models that predict an algorithm's runtime given a problem instance. We discuss the construction of these models and describe techniques for interpreting them to gain understanding of the characteristics that cause instances to be hard or easy. We also present two applications of our models: building algorithm portfolios that outperform their constituent algorithms, and generating test distributions that emphasize hard problems.We demonstrate the effectiveness of our techniques in a case study of the combinatorial auction winner determination problem. Our experimental results show that we can build very accurate models of an algorithm's running time, interpret our models, build an algorithm portfolio that strongly outperforms the best single algorithm, and tune a standard benchmark suite to generate much harder problem instances.

show abstract

Section: :3mentioning

confidence: 99%

Empirical hardness models

Leyton‐Brown

Nudelman

Shoham

2009

J. ACM

View full text Add to dashboard Cite

show abstract

“…Many architectures for SLM-enforcement do exist for multi-tier environments [15,16,17] or single enterprise components [18,19]. Some of these architectures already employ controllers that are able to manage certain aspects of the system without human interaction (self-management, autonomic management).…”

Section: Related Workmentioning

confidence: 99%

Decentralised QoS-Management in Service Oriented Architectures

Schmid¹,

Kroeger²

2008

Distributed Applications and Interoperable Systems

View full text Add to dashboard Cite

Abstract. Traditional hierarchical Service Level Management (SLM) frameworks fail to cope with the challenges imposed by the runtime dynamics of Service Oriented Architectures (SOA). This paper introduces a decentralised management approach that successfully uses emerging selfmanagement techniques to realise a flexible SLM system and presents an architecture that implements this approach. The architecture consists of a modular self-manager framework that provides the basis for component-level and workflow-level management. It provides sensor and effector modules to monitor and manage different classes of applications. Integration with existing SOA components is based on the Service Component Architecture (SCA). The presented framework has been prototypically implemented and is currently evaluated in terms of efficiency and scalability.

show abstract

“…More specifically, online minimization of the response time of an Apache web server by dynamic tuning of the number of maximum clients allowed to be connected simultaneously is described in [17], where hill climbing and fuzzy control techniques are employed. For a database server, online adjustment of multiple configuration parameters using online random and direct search techniques is proposed in [8] to guarantee good performance. For application servers, optimal configurations have also been sought in [21] using off-line experimentation and statistical analysis.…”

Section: Related Workmentioning

confidence: 99%

Self-optimizing block transfer in web service grids

Gounaris

Yfoulis²,

Sakellariou

et al. 2007

Proceedings of the 9th Annual ACM International Workshop on Web Information and Data Management

View full text Add to dashboard Cite

Nowadays, Web Services (WSs) play an increasingly important role in Web data management solutions, since they offer a practical solution for accessing and manipulating data sources spanning administrative domains. Nevertheless, they are notoriously slow and transferring large data volumes across WSs becomes the main bottleneck in such WS-based applications. This paper deals with the problem of minimizing at runtime, in a self-managing way, the data transfer cost of a WS encapsulating a data source. To reduce the transfer cost, the data volume is typically divided into blocks. In this case, response time exhibits a quadratic-like, non-linear behavior with regards to the block size; as such, minimizing the transfer cost entails finding the optimum block size. This situation is encountered in several systems, such as WS Management Systems (WSMSs) for DBMS-like data management over wide area service-based networks, and WSs for accessing and integrating traditional DBMSs. The main challenges in this problem include (i) the unavailability of an analytical model; (ii) the presence of noise, which incurs local minima; (iii) the volatility of the environment, which results into a moving optimum operating point; and (iv) the requirements for fast convergence to the optimal size of the request from the side of the client rather than of the server, and for low overshooting. This paper presents two novel solutions for detecting the optimum block size during data transmission, thus yielding lower response times. The solutions are inspired by the broader areas of runtime optimization and switching extremum control. They incorporate heuristics to avoid local optimal points, and address all the afore-mentioned challenges. The effectiveness and efficiency of the solutions is verified through empirical evaluation in real cases.

show abstract

Generic Online Optimization of Multiple Configuration Parameters with Application to a Database Server

Cited by 23 publications

References 13 publications

Empirical hardness models

Empirical hardness models

Decentralised QoS-Management in Service Oriented Architectures

Self-optimizing block transfer in web service grids

Contact Info

Product

Resources

About