Enabling Interoperable e-Infrastructures with Common Open Standards

E-science projects increasingly require resources in more than one e-Infrastructure, especially when combining high throughput computing and high performance computing concepts in one scientific workflow. Alongside this evolution, adoption of the Open Grid Services Architecture (OGSA) concept (defined by Foster et al. in 2002 [OGSA]) is still slow. While OGSA represents a good architectural blueprint for a wide variety of infrastructures, its scope may be too broad to be realistically implementable for today’s e-Infrastructures. This is perhaps due to the slow emergence of OGSA-* standards and the slow deployment of standard implementations on production e-Infrastructures.
The absence of a realistically implementable standards-based reference model has led to the deployment of numerous different and non-interoperable protocols and interfaces on worldwide e-Infrastructures.
For example, in Europe, the EGEE infrastructure uses gLite middleware while DEISA uses UNICORE. In the U.S., Open Science Grid uses VDT while TeraGrid is based on Globus. Although OGSA aims to facilitate the interoperability of different infrastructures, interoperability between e-Infrastructures requires more precision.
To this end, the Open Grid Forum “Grid Interoperation Now” (GIN) community are working on a so-called “interoperability reference model” (IRM) that is more focused on interoperability between e-Infrastructures than OGSA. The IRM does not replace OGSA, but rather trims its functionality, dropping several parts and refining other parts relevant to the interoperability of e-Infrastructures today. IRM is thus a mid-term milestone towards long-term OGSA conformance in the future and is fundamentally based on the production experiences gained by the GIN community.

Popularity in simplicity

The history of computer science has shown that complex architectures often receive less use than their trimmed-down versions. For instance, the complex SGML is less popular than the smaller and simpler XML. Instead of the seven-layer ISO/OSI model, the four-layer TCP reference model has became the de-facto standard. The same principles can be applied with OGSA by defining a more limited, but more usable, reference model. This is increasingly important in the context of economic constraints, since the IRM can significantly reduce maintenance costs until commercial providers become involved.
We have therefore worked on adoption and refinement of the IRM, based on lessons learned from worldwide GIN interoperation efforts [GIN]. In fact, many parts of the IRM are going to be standardized in the recently formed OGF “Production Grid Infrastructure” (PGI) working group. In short, PGI standards will be based upon numerous already
available open standards but will standardize the missing links between them specifically tuned to the demands of production e-Infrastructures.

IRM in action

By adopting the IRM, we have shown that the WISDOM project can benefit from jointly using the high throughput computing resources of EGEE and the high performance computing resources within DEISA in one scientific workflow [WISDOM]. Users thus benefit from seamless access to a wider variety of resources and services, significantly improving their productivity. At the Supercomputing 2008 conference in Austin, we also have demonstrated a pre-production IRM adoption in the context of the Virtual Physiological Human project. This project actually demonstrated interoperability between TeraGrid, DEISA, and the UK’s National Grid Service by using large-scale HPC simulations.
This satisfies increasing application demands of end users by harnessing worldwide interoperable e-Infrastructures. IRM users also benefit from better access to special
and therefore rare resources, as well as load-balancing that make use of different cycle availability on days/nights across worldwide e-Infrastructures. We are thus looking forward to the standardization of the IRM in the PGI group and its adoption in other world-wide e-Infrastructures so that end-users can truly use a “United Federation of e-Infrastructures” as a tool in daily work.