Adaptive resource scheduling for network services

Recently, there has been considerable interest in providing high performance codes as network services. In particular, high performance network services provide improved performance by executing complex and time-consuming applications (or part of an application) on remote high performance resources. However, since service providers resources are limited, without effective resource scheduling, end-users will not experience performance improvement. In this paper, we propose adaptive resource harvesting algorithms to schedule multiple concurrent service requests within network services. The preliminary results show that our approach can achieve service time improvement up to 40% for a prototypical parallel service.