In the realistic Object-based Storage Device (OSD) workloads, one of the major factors causing I/O bottleneck is I/O request bursty. If the burstiness existing in OSD workloads can be predicted, the bottleneck will be alleviated by establishing a new I/O scheduling mechanism which dynamically allocates the I/O bandwidth based on the predictable traffic. In this paper, we showed that the workload burstiness is pictorially and statistically self-similar which conventional Poisson model is inappropriate to characterize accurately. And we also demo...
