Storage Research for Data-Intensive Applications

19 August 2011: Yusuke Tanimura from AIST, Japan

In this talk, I first introduce myself and my past studies for grid computing briefly. I used to study on a master-slave style grid programming framework (GridRPC), and a grid file system (Gfarm), which provides a global name space and achieves scalable I/O performance by using data access locality. Then I present my current study on the storage resource management. My colleagues and I propose an `advance reservation' concept for the storage access. Not only storage spaces but also I/O components involved with the access, such as disks, storage networks, buffer spaces of the storage servers, and etc, are treated as storage resources, and they can be reserved in the storage system build with our storage software (Papio). From users' view, users can explicitly reserve I/O rate (e.g., MB/s) in access to the storage system, with specifying start and end time of the access. In order to realize the concept, I am studying on an optimized resource allocation method, performance description of storage access, I/O scheduling with flash disks, a Web-services based reservation protocol, an AWS S3 like bucket/object data access interface, and etc. At the end, I also introduce my other interests for scalable storage technologies which support data intensive scientific applications.