Toolbox for Dimensioning Windows Storage Systems

Studies on Windows I/O systems are confined to global performance benchmarking and explanations. Users willing to optimize their I/O systems do not know which tool to use, and up to what point it will effectively represent their applications. In this paper, we describe a toolbox allowing developers and computer architects to better dimension and optimize their I/O systems according to application needs and constraints on single dish architectures. Our toolbox is based on the complementarity of benchmarking and simulation. Its use prevents from having critical I/O performance gaps and helps choosing the optimal I/O strategy. In fact, we observed for the same initial I/O pattern, very frequent dramatic performance drops by a factor three while just changing the access mode or request sizes. Three tools are outlined: (i) a flexible I/O benchmarking tool (ii) an easy-to-use storage parameter extraction tool allowing the explanation of the performance gaps (iii) and an accurate I/O simulator predicting the performance for both real ans synthetic traces with only a 6% mean error