查看文章

Storage researchers and developers strive towards creating systems that store and retrieve data in correct, efficient, available, reliable, durable, fault-tolerant and costeffective manner. Achieving these desirable properties for all I/O requests is a very hard challenge. I/O classification is an effective technique to mitigate this. I/O requests can be categorized based on their properties and better quality of service, performance and reliability be provided to the important I/O classes. This dissertation develops three novel non-invasive I/O classification techniques that work with many different file systems without significant additional implementation effort. Non-invasive techniques that do not require extensive modifications to existing systems face little resistance while making their way into current storage systems.