Making the Grid transparent to users
A key objective of the ongoing EU Grid programme is to make available large-scale, distributed resources capable of solving complex processing problems. The environment, energy, health, transport and industrial design are all likely application areas. At the end of 2003 the Grid infrastructure is already a reality, interconnecting national research networks in Europe and across the world. The next question is - how do researchers access and use this huge resource?
One of largest research undertakings in Europe
Enabling this access is the task of the IST GRIDLAB project, scheduled to deliver its results at the end of 2004. Lead by the Poznan Supercomputer and Networking centre (PSNC) in Poland, the 11 partners in GRIDLAB are working to provide Grid users with a simple and robust environment that allows them to develop applications capable of exploiting the full power and capacity of the Grid.
The project is one of the biggest research undertakings in Europe in the development of application tools and middleware for the Grid environment. It aims to produce a set of application-oriented Grid services such as dynamic resource brokering, monitoring and data management for both end-users and developers.
These services are accessed using the Grid Application Toolkit (GAT), a set of mechanisms (APIs) that provide applications with access to various Grid resources, specific libraries, tools, etc., as well as to the GridLab services themselves. GAT enables applications to make use of whatever resources on the Grid, available at the start of the specific processing task. Thus end-users and application developers can develop and run applications on the Grid without having to know in advance what the runtime environment will provide.
Hiding the complexity of the Grid
"GAT hides the complexity of the Grid from the user," says project coordinator Jarek Nabrzyski of PSNC. "It makes the Grid transparent. Users dont have to worry about which service or resource they are accessing - they use the same API. GAT chooses the best resource available automatically."
While GridLab researchers are making extensive use of specific application frameworks (i.e. Cactus, Triana as application examples for developing GridLab), GAT is designed to be useful for applications and users of all types. It consists of an API, a library and a set of Grid middleware which together allow applications to access Grid resources without the programmer having to learn the details of the underlying Grid middleware or its APIs.
As well as producing the GAT, GridLab researchers are developing and testing Grid applications on real testbeds, constructed by linking the different supercomputers and other resources spanning the globe. So in addition to EU resources, production testbeds are being employed in the US to ensure full interoperability. Testing is carried out by several large user communities, including a European astrophysics network and various multidisciplinary US-funded consortia.
One GridLab service is noteworthy here. The Grid Resource Management System (GRMS) is a service that offers more efficient management of user applications in distributed grid environments. It has already been tested on complex application scenarios that access processing resources in computer centres throughout Europe and the US.
Already demonstrated successfully
GridLab has already demonstrated GAT and its associated services successfully. For example at the Supercomputing 2003 conference (in Phoenix, USA, November 2003), the GAT prototype and underlying GridLab services were shown. It allowed generic applications to migrate across different resources and machine architectures of the GridLab testbed.
In this scenario, users launched production quality Cactus simulations using MPI and Fortran to model the collision in space of two black holes. The purpose of the Cactus application is to study a variety of astrophysical phenomena including black holes, colliding neutron stars, singularities, gravitational waves and similar effects.
Such applications have a very high need for processing power, says Nabrzyski. "Simulating the collision of black holes in space, for example, can require three weeks of computing on a one thousand processor machine. Using the Grid enables us to make use of various distributed processing resources at the same time, and decrease the time required by a factor of three to five."
He again stresses the importance of GAT for providing Grid access to support such processor-hungry applications. "The most important advantage for Grid users is that they have a single tool that allows them to use the Grid in the simplest possible way. That is the big advantage of GAT. People developing their applications for the Grid can take the GAT and they have everything they need."
Poznan Supercomputing and Networking Centre (PSNC)
ul. Noskowskiego 10
Tel. +4861 858 2072
Fax +4861 852 5954
Tara Morris | IST Results