Optimizing the Stretch for Virtual Screening Application on Pilot-agent Platforms on Grid/Cloud by using Multi-level Queue-based Scheduling

Virtual screening has proven very effective on grid infrastructures. We focus on finding platform scheduling policy for pilot-agent platform shared by many virtual screening users. They need a suitable scheduling algorithm at platform level to ensure a certain fairness between users. Optimal criterion used in our research is the stretch, a measure for user experience on the platform. From our latest research (Quang et al., 2013), simulation result and experimentation on real pilot agent platform showed that SPT policy is the best policy in 4 different existing scheduling policies (FIFO, SPT, LPT and Round Robin) for optimizing the stretch. However, research on real grid workload (Medernach, 2005) showed that there are two types of grid user: normal users who submit frequently little jobs to grid and data challenge users who submit occasionally large number of jobs to grid. And SPT policy, in particularly, is not appropriate for data challenge user because they have to wait always normal user. In this paper, we proposed a new policy named SPT-SPT which uses multi-level queue scheduling technique for scheduling in a pilot agent platform. In SPT-SPT policy, the administrator creates two separate user groups in the platform: Normal group and Data Challenge group. Each group has their own task queue in the platform and SPT policy is applied on it. A parameter p (p ε [0,1]), the probability that task queue is chosen to send pilot agent their task, is assigned to one task queue and 1-p for the other one. This policy improves user experience for Data Challenge group and do not impact very much for Normal group.