The paper describes the overall design and implementation of the geographically distributed computing and data storage centre as built by the Nordic Data Grid Facility (NDGF) with the driving goal to serve the needs of the Large Hadron Collider research at CERN. In striving to federate heterogeneous resources in different administrative domains, NDGF chose to create an infrastructure using software services offered by different providers: ARC for computing, dCache for storage, SGAS for usage accounting, and elements of gLite for interoperability with the worldwide LHC Computing Grid.

This article presents a metadata catalog, intended for use in grids. The catalog provides high availability, by replication across several hosts. The replicas are kept consistent using a replication protocol based on the Paxos algorithm. A majority of the replicas must be available in order for the system to function. The data model used in the catalog is RDF, which allows users to create their own name spaces and schemas. Querying is performed using SPARQL. Additionally the catalog can be used as a synchronization mechanism, by utilizing a compare and swap operation. The catalog is accessed using HTTP with proxy certificates, and uses GACL for flexible access control. The performance of the catalog is tested in several ways, including a distributed setup between geographically separated sites.

ARC and gLite are two of the leading production-ready Grid middleware solutions being used by thousands of researchers every day. Even though the middlewares leverage the same technologies, there are substantial architectural and implementation divergences. Today, users face difficulties trying to cross the boundaries of the two systems. The gLite clients have so far not been capable of accessing ARC resources and vice versa. This paper is a follow up on an earlier proposal on how to enable interoperability between these two middlewares. Further, the paper presents a thorough walkthrough of the protocols and steps involved in the submission of a job in grids built up of the two different middlewares.