Section: New Results

Performance Characterization and Optimization of IOs

In high-performance computing environments, parallel file systems provide a shared storage infrastructure to applications. In the situation where multiple applications access this shared infrastructure concurrently, their performance can be impaired because of interference. In [22] , we improve improve performance by alleviating interference effects through a smart I/O scheduler scheduler that organizes and optimizes the applications' requests and adjusts the access pattern to the device characteristics. We apply machine learning techniques to automatically select the best scheduling algorithm for each situation. Our approach improves performance by up to 75

In [33] , we present a new storage device profiling tool that characterizes the sequential to random throughput ratio for reads and writes of different sizes. As we explained previously, several optimizations aim at adapting applications’ access patterns in order to generate contiguous accesses for improved performance when accessing storage devices like hard disks. However, when considering other storage options like RAID arrays and SSDs, the access time ratio between contiguous and non-contiguous accesses may not compensate for these optimizations’ cost. In this scenario, the information provided by our tool could be used to dynamically decide if optimizations are beneficial for performance, which is why we took a particular attention to obtain accurate information in a minimal benchmarking time.