Aurélie Hurault

On deploying scientific software within the GRID-TLSE project

The main goal of the Grid-TLSE project is to design an expert site that provides an easy access to a number of sparse matrix solver packages allowing their comparative analysis on user-submitted problems, as well as on matrices from collections also available on the site. The site provide user assistance in choosing the right solver for its problems and appropriate values for the solver parameters. A computational Grid is used to deal with all the runs arising from user requests. After an overview of the project, we discuss on some of the main difficulties we face in Grid-TLSE: considering the amount of softwares that will be available, facilitating to facilitate their deployment and their exploitation over a Grid is crucial. We make use of an abstract meta-data based description of solvers and have introduced the concept of scenario, i.e. a high-level user assistance description that provides ways of discovering automatically the most appropriate solvers based on their description. The scenarios are used for generating the dynamic workflows that will be executed over the Grid and take advantage of some of the features available in the DIET Grid middleware. We also report on the introduction of a semantic-based component model for efficient service trading.

A Grid-Aware Web Portal with Advanced Service Trading for Linear Algebra Calculations

Due to the rapid growth of the Internet, there has been a rising interest in using the Web as an interface to develop various applications over computational Grid environments. The purpose of this work is to develop a Grid-aware Web interface for linear algebra tasks with advanced service trading. Developing efficient and portable codes, requires users to face parallel computing and programming and to make use of different standard libraries, such as the BLAS [1], LAPACK [2] and ScaLAPACK [3] in order to solve computational tasks related to linear algebra. For this purpose, a scientific computing environment based on a Web interface is described that allows users to perform their linear algebra tasks without explicitly calling the above mentioned libraries and softwarep tools, as well as without installing any piece of software on local computers: users enter algebraic formula (such as in Matlab or Scilab [4]) that are evaluated for determining the combinations of services answering the user request. Services are then executed locally or over the Grid using the Distributed Interactive Engineering Toolbox (DIET) [5] middleware.

On the Easy Use of Scientific Computing Services for Large Scale Linear Algebra and Parallel Decision Making with the P-Grade Portal

Scientific research is becoming increasingly dependent on the large-scale analysis of data using distributed computing infrastructures (Grid, cloud, GPU, etc.). Scientific computing (Petitet et al. 1999) aims at constructing mathematical models and numerical solution techniques for solving problems arising in science and engineering. In this paper, we describe the services of an integrated portal based on the P-Grade (Parallel Grid Run-time and Application Development Environment) portal (http://www.p-grade.hu) that enables the solution of large-scale linear systems of equations using direct solvers, makes easier the use of parallel block iterative algorithm and provides an interface for parallel decision making algorithms. The ultimate goal is to develop a single sign on integrated multi-service environment providing an easy access to different kind of mathematical calculations and algorithms to be performed on hybrid distributed computing infrastructures combining the benefits of large clusters, Grid or cloud, when needed.

Intelligent service trading and brokering for distributed network services in GridSolve

One of the great benefits of computational grids is to provide access to a wide range of scientific software and a variety of different computational resources. It is then possible to choose from this large variety of available resources the one that solves a given problem, and even to combine these resources in order to obtain the best solution. Grid service trading (searching for the best combination of software and execution platform according to the user requirements) is thus a crucial issue. Trading relies on the description of available services and computers, on the current state of the grid, and on the user requirements. Given the large amount of services that may be deployed over a Grid, this description cannot be reduced to a simple service name. In this paper, a sophisticated service specification approach similar to algebraic data types is combined with a grid middleware. This leads to a transparent solution for users: they give a mathematical expression to be solved, and the appropriate grid services will be transparently located, composed and executed on their behalf.

Introduction of a Grid-aware portlet for numerical calculations

Grid portals are one of the most popular user interfaces to Grids. Grid portals build upon the familiar Web portal model offer to virtual communities of users a single access point to computational or data resources. P-GRADE is a Grid portal solution that allows users to manage the whole life-cycle for executing a parallel application in the Grid. The purpose of this article is to introduce the structure of a Grid-aware portlet for linear algebra calculations based on the P-GRADE portal. To accomplish this goal, the portlet provides the seamless bridge between the linear algebra calculations and various linear algebra software environments (middlewares, tools, parallel programming techniques, linear algebra libraries) deployed over a grid. The portlet GUI (Graphical User Interface) is lightweight and uses standard web technologies. Moreover, since today smartphone are ubiquitous, we propose to provide an easy and adapted way to monitor the portlets operation from a mobile device and illustrate its practical use.

On the diversity of asynchronous communication

Asynchronous communication is often viewed as a single entity, the counterpart of synchronous communication. Although the basic concept of asynchronous communication is the decoupling of send and receive events, there is actually room for a variety of additional specification of the communication, for instance in terms of ordering. Yet, these different asynchronous communications are used interchangeably and seldom distinguished. This paper is a contribution to the study of these models, their differences, and how they are related. In this paper, the variety of point-to-point asynchronous communication paradigms is considered with two approaches. In the first and theoretical one, communication models are specified as properties on the ordering of events in distributed executions. In the second and more practical approach that involves composition of peers, they are modeled with transition systems and message histories as part of a framework. The described framework enables to model peer composition and compatibility properties. Besides, an implemented tool chain based on the TLA+ formalism and model checking is also proposed and illustrated. The conformance of the two approaches is highlighted. A hierarchy is established between the studied communication models. From the execution viewpoint, it completes existing work in the area by introducing more asynchronous communication models and showing their differences. The framework is shown to offer abstract implementations of the communication models. Both the correctness and the completeness of the descriptions in the framework are studied. This reveals necessary restrictions on the behavior of the peers so that the communication models are actually implementable.

Selecting linear algebra kernel composition using response time prediction

Numerical linear algebra libraries provide many kernels that can be composed to perform complex computations. For a given computation, there is typically a large number of functionally equivalent kernel compositions. Some of these compositions achieve better response times than others for particular data and when executed on a particular computer architecture. Previous research provides methods to enumerate (a subset of) these kernel compositions. In this work, we study the problem of determining the composition that yields the lowest response time. Our approach is based on a response time prediction for each candidate combination. While this prediction could in principle be obtained using analytical and/or empirical performance models, developing accurate such models is known to be challenging. Instead, we define a feature space that captures salient properties of kernel combinations and predict response time using supervised machine learning. We experiment with a standard set of machine learning algorithms and identify an effective algorithm for our kernel composition selection problem. Using this algorithm, our approach widely outperforms the strategy that would consist in always using the simplest kernel composition and is often close to the fastest kernel compositions among those evaluated. We quantify the potential benefit of our approach if it were to be implemented as part of an interactive computational tool. We find that although the potential benefit is substantial, a limiting factor is the kernel composition enumeration overhead. Copyright

Enabling Large-Scale Linear Systems of Equations on Hybrid HPC Infrastructures

Scientific research is becoming increasingly dependent on the large-scale analysis of data using High Performance Computing (HPC) infrastructures. Scientific computing aims at constructing mathematical models and numerical solution techniques for solving problems arising in science and engineering. The solution of linear system of equations lies at the heart of most calculations in scientific computing. HPC infrastructures with many-core and graphics processing unit (GPU) challenges, Cloud and Grid technologies and e-infrastructures are currently offering interesting opportunities for solving large-scale linear system of equations. In this article, a second-generation of our Web portal for Scientific Computing is introduced based on a hybrid HPC infrastructure that provides predictable optimal execution and scales from a single resource to multiple resources. After analyzing the synergies and the complementarities of the different computing platforms, we argue for an architecture that combines the benefits of these technologies.

Mechanized Refinement of Communication Models with TLA

In distributed systems, asynchronous communication is often viewed as a whole whereas there are actually many different interaction protocols whose properties are involved in the compatibility of peer compositions. A hierarchy of asynchronous communication models, based on refinements, is established and proven with the TLA