Catálogo de publicaciones - libros

Before 1990, the world wide network was almost entirely unknown outside the universities and the corporate research departments. The common way of accessing the Internet was via command line interfaces such as , , or popular Unix mail user agents like , , , or . The usual access to information was based on peer-to-peer email message exchange which made the every day information flow slow, unreliable, and tedious. The advent of the has revolutionised the information flow though the Internet from obsolete message passing to world wide Web page publication. Since then, the Internet has exploded to become an ubiquitous global infrastructure for publishing and exchange of (free) digital information.

The mostly used attempt to define Grid computing [77] is through an analogy with the electric power evolution around 1910. The truly revolutionary development was not the discovery of electricity itself, but the electric power grid that provides standard, reliable, and low cost access to the associated transmission and distribution technologies. Similarly, the Grid research challenge is to provide standard, reliable, and low cost access to the relatively cheap computing power available nowadays.

Existing parameter study tools provide support to specify value ranges for application parameters of interest, e.g. by means of external scripting languages [5], or through graphical annotation of input files [197]. All these approaches, however, force the user to export the application parameters to global input files or program arguments, which often requires undesired source code adaptation for using the tool. Additionally, there are no tools that combine the experiment specification and management with cross-experiment performance analysis. All currently existing performance tools are restricted to single experiment analysis, which is not enough for efficient application performance tuning, that is inherently a multi-experimental process.

We designed ZENTURIO [140, 144] as a tool to automatically generate and conduct large number of experiments in the context of large scale performance and parameter studies on cluster and Grid architectures. ZENTURIO uses the ZEN language presented in Chapter 3 to specify a large set of performance and parameter study experiments in a compact and user friendly manner. Thereafter, it automatically generates, conducts, and analyses the performance and output data through a distributed service-oriented Grid architecture shielded from the end-user by means of a graphical User Portal. ZENTURIO systematically organises the performance and output data produced by all experiments into a well-defined Experiment Data Repository for post-mortem analysis.

As applications get larger and more complex, the use of software tools becomes vital for tuning application parameters, identifying performance leaks, or detecting program defects. Extensive efforts within academia and industry over the last decade have resulted in a large collection of tools for practical application engineering. Available tools of broad interest include program source and structure browsers, editors, static program analysers, performance predictors, optimisation compilers, execution control and monitoring environments, sequential and parallel debuggers (providing deadlock detection and deterministic message replay mechanisms), data and execution visualisers, performance analysers, or various program tracers.

We introduced in Chapter 4 the ZENTURIO experiment management tool for multi-experimental performance and parameter studies of parallel applications. To achieve this goal, ZENTURIO performs an automatic sweep of the entire parameter space defined using the ZEN directive-based experiment specification language described in Chapter 3.

With the emergence of Grid computing that aggregates a potentially unbounded number of resources, new classes of applications such as workflows and parameter studies are of increasing interest to the scientists. The parameter space of such large scale Grid applications can easily achieve rather huge dimensions for which the exhaustive parameter sweep performed by ZENTURIO is no longer a feasible solution. In general, a complete parameter sweep gives useful detailed insight on the application behaviour but also produces vast amounts of data that are irrelevant for further studies. Often the user’s ultimate goal is to find parameter combinations that a certain application behaviour, such as a performance metric or an output result. Such optimisation problems are known as [85] and require advanced heuristics to find approximate or reasonably good solutions.

Workflow modeling is a well established area in computer science that was strongly influenced by business process modeling work [187]. Recently, the Grid community has generally acknowledged that orchestrating existing software applications implemented as Grid services in course grain workflows represents an important class of applications that matches the loosely coupled Grid model and, therefore, can benefit from being executed in distributed Grid infrastructures. Similarly, in order to efficiently harness the computational resources provided by the Grid, existing monolithic scientific applications are currently being re-engineered and decomposed in a set of atomic activities orchestrated in a loosely coupled scientific workflow [58, 133].

The work presented in this monograph is centred around six different research fields: experiment management, performance studies, parameter studies, tool integration, scheduling, and scientific workflows. In the following sections we outline the most relevant related work in each of these areas.

In this section we conclude this research monograph by summarising our main contributions in the areas of experiment management, scheduling, tool integration, and scientific workflow management in Grid computing.