Service oriented architectures allow modelling engines to be hosted over the Internet abstracting physical hardware configuration and software deployments from model users. Many existing environmental models are deployed as desktop applications running on user's personal computers (PCs). Migration to service - based modelling centralizes the modelling functions to service hosts on the Internet . Users no longer require high-end PCs to run models and model updates encapsulating science advances can be disseminated more rapidly by hosting the modelling functions centrally via an Internet host instead of requiring software updates to user's PCs . In this paper we present the Cloud Services Innovation Platform (CSIP), an Infrastructure -as -a -Service cloud application architecture , used to prototype development of distributed and scalable environmental modelling services. CSIP aims to provide modelling as a service to support both interactive (synchronous) and batch (asynchronous) modelling. CSIP enables c loud-based computing resources to be harnessed for both new and existing environmental models supporting the disaggregation of work into subtasks which execute in parallel using a scalable number of virtual machines. This paper presents CSIP 's implementation using the RUSLE2 model as a prototype model. RUSLE2 model service benchmarks are presented to demonstrate performance gains from using cloud resources. We also provide benchmarks for virtualization overhead observed using popular virtual machine hypervisors and demonstrate how application profile characteristics significantly impact performance when virtualized.
Available at: http://works.bepress.com/wes-lloyd/3/