Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Computational Science, Engineering & Technology Series
PARALLEL, DISTRIBUTED AND GRID COMPUTING FOR ENGINEERING
Edited by: B.H.V. Topping, P. Iványi
Performance Trees: A Query Specification Formalism for Quantitative Performance Analysis
W.J. Knottenbelt, N.J. Dingle and T. Suto
Department of Computing, Imperial College London, United Kingdom
W.J. Knottenbelt, N.J. Dingle, T. Suto, "Performance Trees: A Query Specification Formalism for Quantitative Performance Analysis", in B.H.V. Topping, P. Iványi, (Editors), "Parallel, Distributed and Grid Computing for Engineering", Saxe-Coburg Publications, Stirlingshire, UK, Chapter 9, pp 165-198, 2009. doi:10.4203/csets.21.9
Keywords: performance trees, PIPE2, generalised stochastic Petri nets, stochastic modelling.
Stochastic modelling and associated performance analysis techniques allow engineers to understand and quantify the fundamental performance characteristics of complex distributed systems, both before and after implementation. This can help to avoid the unforeseen performance problems that so often plague real-life systems. In this context, it is a major - albeit largely unaddressed - challenge to find a way to specify complex performance queries on models of systems that is simultaneously accessible, rigorous and expressive.
This paper describes our attempts to address this challenge through the development of Performance Trees [2,3,6-8], a formalism for the graphical specification of complex performance queries on stochastic models. Performance Trees are designed to be accessible by providing more intuitive query specification (including structured natural language-based specification techniques ), expressive by being able to reason about a broader range of concepts than current alternatives such as stochastic logics, extensible by supporting additional user-defined concepts through macro functionality, and versatile through their applicability to multiple modelling formalisms including stochastic Petri nets, stochastic process algebras and queueing networks. The formalism is backed up by a rigorous theoretical framework that defines the syntax, typing and quantitative semantics of operators .
Prototype tool support is key to encouraging the adoption of any new formalism. Here this is implemented in the form of a module for the PIPE2 Petri net tool [1,3,5], which provides Performance Tree query design capabilities through both a graphical user interface and a structured natural language query builder. Query evaluation is supported by a set of integrated parallel and distributed analysis tools, which are hosted on a dedicated cluster. The application of Performance Trees is demonstrated in the context of a case study of an online transaction system. We also show how they can be applied to concepts found in Service Level Agreements (SLAs), and propose a unified PT-based framework for the specification, compliance prediction and online monitoring of SLAs .
purchase the full-text of this chapter (price £20)