A Virtualization Pro-middleware For Grid Computing

Tools that empower end-users to take full advantage of the grid infrastructure withoutrelying on support teams are essential to fulfill the original grid vision of computingpower as a utility accessible to all. Across a broad set of industry sectors, theproblems scientists and engineers need to solve are growing in size and complexity.This growth is by far outpacing the hardware advancement delivered by Moore's Law.Most of the scientists and engineers who want to solve computationally intensiveproblems have not adopted the grid computing approach, because its use requires agroup of knowledgeable, dedicated specialists. So we still need to develop tools tofacilitate end-users to directly run their applications on grid by themselves withoutdepending on expertise of support teams. A much wider variety of applications andlarge community of users can get benefit once grid computing technologies becomeeasier to use and more sophisticated.This thesis aims to present a virtualization pro-middleware Users-Grid, which sitsbetween the grid middleware and user applications. It provides a high-levelabstraction layer to grid applications and hides the intricacies of the grid middleware.Further-more, it insulates users and developers from the complexities of themiddleware (Globus) infrastructure. The core of the Users-Grid framework is apersonal submission agent that automatically performs the steps involved in jobsubmission. An increasing number of research groups have been implementingmiddleware, libraries, and tools that allow cooperative use of geographicallydistributed resources. A legacy application can only be used through its specificinterface, and cannot be modified. In many cases, the cost of rewriting a legacyapplication would be prohibitive. Using middleware approach in this scenario is theonly option and it's the main idea of Pro-middleware approach of Users-Grid. Matlab integrates the rapid prototyping of numerical computations with high-levelprogramming. Because interactive desktop science & engineering tools (e.g. Matlab,Mathematica, IDL, Maple, and Python) cannot be used on grids, so lots of time andeffort is spent moving problems from desktops to grids. But, there is now anopportunity to bring the power of grid to today's and tomorrow's users, byeliminating the last hurdle: the need to manually re-program models for gridcomputing initially prototyped using desktop tools. Users-Grid minimizes time-to-solution by allowing very high-level languages (VHLLs) tools to run on the grid.Most large organizations view the evolution of their information technologyinfrastructure as evolutionary, not revolutionary. This translates into the requirementto support heterogeneous configurations in which proprietary legacy systems,supporting operational applications, interact with new applications. This need isfulfilled by the Pro-middteware approach of Users-Grid. It is to note that no softwarepackage or tool is currently available to enable Matlab to run on the grid environment.This thesis also presents a prototype implementation and discusses severalexperimental results. The experimental results show that Users-Grid performs farbetter than serial Matlab. And when compared with other software packages onlyMatlab*G comes closer to the efficiency provided by Users-Grid.