Distributed file systems: concepts and examples
ACM Computing Surveys (CSUR)
Storage area network essentials: a complete guide to understanding and implementing SANs
Storage area network essentials: a complete guide to understanding and implementing SANs
High Performance Cluster Computing: Programming and Applications
High Performance Cluster Computing: Programming and Applications
Grid Computing: Making the Global Infrastructure a Reality
Grid Computing: Making the Global Infrastructure a Reality
Storage Networking Fundamentals: An Introduction to Storage Devices, Subsystems, Applications, Management, and File Systems (Cisco Press Fundamentals)
The Holy Grail of Network Storage Management
The Holy Grail of Network Storage Management
Hi-index | 0.01 |
This paper describes and analyzes a new architecture for file systems in which `metadata', lock control, etc., are distributed among diverse resources. The basic data structure is a segment, viz. a logical group of files, folders, or other objects. The file system requires only one root, and can be non-hierarchical without a complete tree structure within segments. For `embarrassingly parallel' data distributions, scalability is trivially perfect for all N,where N is the number of servers. Even for random file access, a new extreme statistical mechanics is used to show that data I/O is `perfectly' scalable with probability 1, with degradation from perfect scaling that is small and bounded by f ln N/ ln (ln N). Here f is the fraction of data that is metadata. In contrast, earlier solutions degrade much faster, like Nf. No structural changes in classical metadata are required.