gLite is an ambitious effort to create Grid middleware for a range of scientific applications, based primarily on "best of breed" solutions from other Grid projects. It is now being comprehensively tested and some parts are already in use under production conditions.
Tout glisse sur la Grille grâce à gLite
gLite désigne un ambitieux programme visant à créer un intergiciel de Grille pour diverses applications scientifiques de natures différentes, en exploitant préférentiellement les "meilleures solutions" élaborées dans d'autres projets de grilles. Ce travail de développement est un exemple parfait de collaboration internationale, car les partenaires principaux y sont entrés dès le début de la conception. Actuellement, on soumet l'intergiciel à des essais complets et certaines de ses parties sont déjà en utilisation, dans les conditions de production, en particulier pour les essais d'exploitation de la Grille de calcul du LHC.
Any Grid infrastructure consists of three basic building blocks: on the one side, the underlying infrastructure (or fabric) providing computing and storage resources; on the other, the users with their applications, wanting to use the resources; and bringing the two together, the so-called "middleware". Aptly named, this software typically consists of a stack of different modules, which act collectively as an intermediary, hiding the multiple parts and detailed workings of the Grid infrastructure from the user. The Grid thus appears as a single, coherent, easy-to-use resource, in which the middleware ensures that the resources are used as efficiently as possible and in a secure and accountable manner.
Many national and international projects are working in the rapidly evolving field of Grid computing, and there are diverse technologies currently available. Recent efforts have focused on closer collaboration between projects to find the best solutions and ensure that viable standards emerge and evolve - a trend that should help to make Grid technology widely accessible to a larger user community.
It is in this spirit that the Enabling Grids for E-science (EGEE) project launched its gLite middleware initiative in April 2004. Rather than trying to build a new middleware stack from scratch, the gLite design team included representatives from existing middleware providers (AliEn, Condor, EDG, Globus, etc), as well as other key stakeholders such as people from the Grid-operations activity in EGEE, and partners from related projects, such as Open Science Grid (OSG, see CERN Courier September 2005 p19) in the US. gLite components were selected based on a thorough analysis of requirements from the pilot application areas in EGEE, biomedicine and high-energy physics.
The gLite middleware is envisaged as a modular system, to allow users to tailor the system to their specific needs by deploying the services they require, rather than using the whole system. The Grid services of gLite follow a service-oriented architecture (SOA), making it easier to connect the software to other Grid services. It will also help users to comply with upcoming Grid standards, for instance the Web Service Resource Framework (WSRF) from OASIS and the Global Grid Forum's Open Grid Service Architecture (OGSA).
Distributed under an open-source licence, gLite is mainly based on middleware from EGEE's predecessor, the European DataGrid (EDG) project, and makes use of components from other ongoing middleware projects (see "gLite services"). Using existing middleware components developed by many different projects worldwide means that gLite inherits code from other sources. The middleware team in EGEE re-engineers and integrates the different components, which are often written in many different programming languages, into a coherent software suite. Each line of code is measured by how well it complies with coding guidelines and standards defined for each language. Individual units or functions are tested to make sure that they do what they are supposed to do. Functional testing is performed on a distributed test-bed at CERN, the CCLRC Rutherford Appleton Laboratory, NIKHEF and Imperial College, London.
gLite is being developed in two main cycles with incremental releases in between. On 4 April 2005 gLite 1.0 was released to be certified by the operations team, which gathers experience in running middleware on a large-scale production-level service. After repeating the functional tests on an independent test-bed, the middleware is subjected to stress tests. During this certification step, the middleware is also tested on different "flavours" of operating systems used in the EGEE Grid resource centres. Currently this is mainly Scientific Linux (see CERN Courier September 2005 p18), but increasingly other architectures are being requested.
Once the middleware is certified, it is deployed on a preproduction test-bed, which includes 12 sites in Europe and the US that reflect the different sizes and usage policies of sites in the wider EGEE infrastructure. gLite is then exposed to some of EGEE's pilot applications, to test the middleware under real conditions. However, this is not easy, since the pilot applications typically have to access data stored on the production Grid service, thus requiring access to both the production and the pre-production services.
Several components of the gLite middleware are already being incorporated into production use, such as information services in clinical decision support systems and in medical imaging. The LHC Computing Grid (LCG) also makes use of a number of gLite services for its service challenges. LCG will have to deal with huge amounts of data - in excess of 10 PB per year - and so data and service challenges are ongoing to test the infrastructure (CERN Courier June 2005 p15). These challenges also affect the development of gLite by prioritizing components of the middleware that need to be ready for a certain challenge. For example, the third LCG service challenge, which started on 26 June, includes a data-throughput test in which the file-transfer service (FTS) of the gLite middleware is used.
The FTS formed part of the incremental release gLite 1.2, released in July, which also included a new version of the virtual-organization membership system (VOMS) and further Condor support of the computing elements. gLite 2.0, which will be released at the end of the year, will focus on secure services, increased robustness and ease of installation, as well as further developments of the core services, incorporating experience gained in the data and service challenges.
Further reading
For further information see www.glite.org.
Author:
Compiled by Hannelore Hämmerle and Nicole Crémel