Practical Considerations in Cloud Utilization for the Science Gateway nanoHUB.org
Date of this Version2011
nanoHUB.org is arguably the largest online nanotechnology user facility in the world. Just between July 2010 and June 2011 it served 177,823 users. 10,477 users ran 393,648 simulation jobs on a variety of computational resources ranging from HUBzero-based virtual execution hosts for rapid, interactive runs as well as grid-based resources for computationally-intense runs. We believe that as such our users experience a fully operational scientific “cloud”-based infrastructure even though it is not using “standard” computational cloud infrastructures such as EC2. In this paper we explore the use of standard computational cloud-based resources to determine whether they can deliver satisfactory outcomes for our users without requiring high personnel costs for configuration. In a science gateway environment, the assignment of jobs to the appropriate computational resource is not trivial. Resource availability, wait time, time to completion, and likelihood of job success must all be considered in order to transparently deliver an acceptable level of service to our users. In this paper, we present preliminary results examining the benefits and drawbacks of utilizing standard computational cloud resources as one potential venue for nanoHUB computational runs. In summary we find that cloud resources performed competitively with other grid resources in terms of wait time, CPU usage, and success in a multiple job submission strategy.
Nanoscience and Nanotechnology