Craig Mattocks

Middleware alternatives for storm surge predictions in Windows Azure

Cloud computing is a resource of significant value to computational science, but has proven itself to be not immediately realizable by the researcher. The cloud providers that offer a Platform-as-a-Service (PaaS) platform should, in theory, offer a sound alternative to infrastructure-as-a-service as it could be easier to take advantage of for computational science kinds of problems. The objective of our study is to assess how well the Azure platform as a service can serve a particular class of computational science application. We conduct a performance evaluation using three approaches to executing a high-throughput storm surge application: using Sigiri, a large scale resource abstraction tool, Windows Azure HPC scheduler, and Daytona, an Iterative Map-reduce runtime for Azure. The differences in the approaches including early performance measures for up to 500 instances are discussed.

Executing Storm Surge Ensembles on PAAS Cloud

Cloud computing services are becoming increasingly viable for scientific model execution. As a leased computational resource, cloud computing enables a computational modeler at a smaller university to carry out sporadic large-scale experiments, and allows others to pay for CPU cycles as needed, without incurring high maintenance costs of a large compute system. In this chapter, we discuss the issues involved in running high throughput ensemble applications on a Platform-as-a-Service cloud. We compare two frameworks deploying and running these applications, namely Sigiri and MapReduce. We motivate the need for a pipelined architecture to application deployment, and discus a couple of methodologies to balance the loads, minimize storage overhead, and reduce overall execution time.