Apache Airavata as a Laboratory

Science gateways are more than user interfaces to computational grids and clouds. Gateways are middleware in their own right, providing flexible, lightweight federations of heterogenous collections of computing resources (such as campus clusters, supercomputers, computational clouds), all of which remain challenges for many alternative middleware approaches. Gateways also are notable for providing science application-centric interfaces to computing resources rather than resource-centric views. An important challenge for science gateway research is to generalize specific science gateway strategies, defining a reference architecture that emcompasses major gateway capabilities while enabling implementation flexibility. Such a reference architecture should also enable "platform as a service" approaches that provide hosted versions of common gateway capabilities. In this paper, we summarize the Apache Airavata software system as a candidate reference architecture for science gateways. We propose the use of a component-based architecture to encompass major gateway capabilities (such as metadata management, meta-scheduling, execution management, and messaging). We examine the messaging system component in this abstract architecture in detail and describe its re-implementation and validation using third party messaging system software to replace a custom-built messaging system. Besides the operational validation of this specific component, we infer a preliminary validation of the overall architecture. The flexibility of component implementations within an overall architecture is essential as it allows gateway middleware to be the subject of distributed computing research for its own sake while also ensuring that we don't get locked into less than optimal implementations for gateway operations.