Flexible Advance Reservation Impact on Backfilling Scheduling Strategies

Advance reservation mechanism was introduced in grid environments to provide time quality of service requirements for time critical applications. Also there are applications that need resource coordination namely co-allocation and workflow, which benefits from this capability. Contemporary trend in grid computing is toward flexibility, because of its advantages in advance reservation such as increasing accept rate. This mechanism should be supported by local scheduler which is responsible for normal jobs. Using backfilling methods with FCFS priority for scheduling normal jobs is the dominant approach. Therefore well investigation of performance impact of flexible Advance Reservation on backfilling scheduling is essential. In this paper incorporation of flexible Advance Reservation in major backfilling policies is investigated, regarding workload parameters impact on well known performance metrics. Here also the impact of increasing inaccuracy of user estimated job runtime is studied. Our experimental results indicate that aggressive backfilling algorithm is biased toward AR related performance metrics but conservative method has better job performance. Although more flexibility in advance reservation would improve AR performance but its impact on degrade of job performance metrics is noteworthy. Both parameter types' rates are less in conservative method. Moreover, usually increasing inaccuracy interestingly has opposing effect on backfilling methods.