Providing Scientific Software as a Service in Consideration of ServiceLevel Agreements

Software-as-a-Service (SaaS) is an interesting concept for providing scientific applications. Service providers in this field have to either run their own data centre or to outsource the applications to a third party provider. In both cases, they want to optimally utilize resources without violating service level agreements (SLAs) negotiated with their customers. This work analyzes the properties of scientific applications. We found that there are essential differences between scientific and business applications. We also build up an own cloud and analyze Gromacs 4, a common scientific application. This shows that for fulfilling an SLA, it is often possible to provide different configurations, using either a few high performance machines or multiple smaller machines. Furthermore, experiences of the performance behavior gained in physical environments must not necessarily apply in virtual environments. Fulfilling SLAs also depends significantly on the mapping of the virtual machines to physical hosts.