A SPECIFIC APPLICATION OF GRID

A specific application of Grid

A specific application of Grid

Introduction

A computational Grid is a large-scale distributed system that supports integration of different computational resources from different administrative domains [1]. With the advent of Grid and related technologies, scientists and engineers are building more and more complex applications to manage and process large data sets, or to implement system models on distributed resources. Some applications are translated in large amounts of individual tasks needed to obtain the answer to the problem being considered. This is the case of parameter sweep problems, which arise naturally in several scientific and engineering fields. In these problems, we have a system defined by a set of independent parameters. The behavior of the system can be determined by varying these parameters. This produces a set of tasks that can be independently computed. Depending on the size of the parameter space, the number of tasks can be very large. It has been shown that the Grid approach is very efficient in dealing with these kinds of problems [2] and [3].

The above mentioned scientific or engineering application scenarios require means for composing and executing complex workflows. The workflow concept is concerned with the automation of procedures where, to achieve an overall goal, data are passed between participants according to a defined set of rules. In this context, the Grid programming environment GRID superscalar (GRIDSs) is able to implement and handle dynamic workflows [4]. In these kinds of applications, users must be supported in finding and keeping track of the available resources, and on the state of the processes running on the Grid. To address these objectives, it is necessary to collect systematically information on the present, and sometimes on the past, status of resources or processes. This is nothing but a process of monitoring. In addition, access to application level steering is a desirable feature. According to [5], only AutoPilot [6] and Mercury [7] incorporate some kind of application oriented steering at the Grid level.

Background

Monitoring is classically defined as the collection, interpretation and display of information concerning the status of the hardware or software system of interest [8]. For instance, to produce parallel applications that perform well on today's parallel architectures, different tools for collecting and analyzing data do exist. As representative tools of this kind we have Tau [9], Paradyn [10], or MATE [11], for a review see [12].

In a Grid, its information system ( Grid Information System or GIS) is in charge of monitoring the status of the resources forming the Grid. However, monitoring is also of key interest for job scheduling, data replication, accounting, data analysis, or for the optimization of applications [13]. Monitoring the status of an application in a Grid environment can be carried out in principle from the information provided by the Grid resource management system. However, in this form, we only can get information about individual tasks. This is not a practical approach when our application handles tens or hundreds of thousands ...