European Middleware Initiative Registry, a solution for federated online services

Positive feedback from the EMIR pilot service preliminary results
Today a federation of online service, such as Distributed Computing Infrastructures (DCI), represents an indispensable tool for scientific research.

In fact, science collaborations increasingly deal with large scale international projects, based on heterogeneous computing services, handling not only the distributed data sets and the processing applications, but also the management technologies used by the participating infrastructures. Such infrastructures are composed of many independent services managed by autonomous service providers, evolving the older model of monolithic, homogeneous resources. In such an environment, the discovery of available services is therefore crucial both for users and other services.

Many existing approaches rely on a centralized service registry and this presents a limitation for use within a federated environment. A solution is required for the flexible, agile, federated operational model of cooperating yet autonomous service providers on which are based the DCI, such as the European Grid Initiative (EGI). The European Middleware Initiative Registry (EMIR) aims to overcome such limitations with a de-centralized approach that supports both hierarchical and peering topologies enabling autonomous domains to collaborate in a federated infrastructure.

Services are grouped with other services managed by the same organization in groups called Domains which can be organized in a hierarchical structure or as a federation. EMIR provides two main components: the Domain Service Registry (DSR) and Global Service Registry (GSR). A service periodically sends information about itself to the DSR which may re-publish this information to a parent DSR/GSR. The GSR replicates this information to the other GSRs within the federation.

A pilot service has recently been deployed to evaluate EMIR under real-world conditions with a large-scale distribution in a controlled environment. The topology used for the pilot service was a simple hierarchy, with a DSR deployed at the site-level and a single GSR to which all the DSRs register. As the pilot service relied on voluntary contributions of system administrators, the deployment overhead was reduced by extracting real service records for the site from the existing site-level information service.

“The EMIR pilot service has been a success! - declares Laurence Field, the coordinator of the pilot service - The initial feedback from the system administrators has been very positive.”

Ten DSRs located at ten sites in ten countries spanning five continents contributed to the pilot service: five sites in Europe1 and five sites world-wide2. In order to verify the pilot service, a simple visualization was created by querying the GSR service records to obtain the name of the site and displaying this geographically using Google maps.

“There is now a DSR installed on every continent, except Antarctica, and the infrastructure has been stable from the beginning. - Laurence concludes. - Performance tests have shown that show that the average query response time for 800 concurrent client requests for a registry containing 10,000 service records was around 21ms. More detailed performance tests, especially relating to the replication of service records between the GSRs are planned, however, based on the preliminary results, EMIR is premising to be a good solution for service discovery in a strongly federated environment.”