Storage I/O has become a performance bottleneck for large-scale computation and storage systems. Characterization of I/O network traffic has been the observed self-similarity nowadays. In this paper, basing on the reviews of the origins of this self-similarity, we study how Stochastic Petri Net (SPN) models can be exploited for self-similar I/O behaviors of virtual storage systems. We then present the SPN models and simplifying processes of file I/O and multimedia I/O for origination node, destination node and communication link. Finally, we propose the self-similarity simulation methodologies. Analysis results show that without self-similar I/O scenario outperforms self-similar I/O scenario both for file I/O and multimedia I/O, and show how the ratio of multimedia/file I/O and the queue capability can affect the observed storage I/O self-similarity.
展开▼
