The role of commercial cloud computing as a source of scalable compute power for science gateways is an area of ongoing investigation. As part of this effort, we are exploring the practicality of cloud bursting to a commercial provider for the CIPRES Science Gateway (CIPRES), a highly accessed gateway that delivers compute resources to users across all fields of biology. CIPRES provides browser and RESTful access to popular phylogenetics codes run on large computational clusters. Historically, CIPRES has submitted compute-intensive jobs to clusters provided through the NSF-funded XSEDE project. An ongoing issue for CIPRES is whether compute time available on XSEDE resources will be adequate to meet the needs of a large and growing user base. Here we describe a partnership with Internet2 to create infrastructure that supports CIPRES submissions to compute resources available through a commercial cloud provider, Amazon Web Services (AWS). This paper describes the design and implementation of the infrastructure created, which allows users to submit a specific subset of CIPRES jobs to V100 GPU nodes at AWS. This new infrastructure allows us to refine and tune job submissions to commercial clouds as a production service at CIPRES. In the short term, the results will speed the discovery process by allowing users greater discretionary access to GPU resources at AWS. In the long term, this infrastructure can be expanded and improved to submit all CIPRES jobs to one or more commercial providers on a fee-for-service basis.
