Changing Requirements for Distributed File Systems in Cloud Storage

File systems typically have centralized metadata servers that present performance bottlenecks as concurrent users and system size increase. These are unique challenges for distributed file systems. Cloud storage systems often store large unstructured content, and the streaming write access patterns typical of such systems allows for optimizations that cannot be made in traditional file systems. A new technique that adapts principals from NOSQL and object storage paradigms - and uses information dispersal for both underlying storage and metadata - provides a viable solution for streaming write access patterns. This technique allows for distributed writes, no single point of failure, scalability of both system size and concurrent clients, and limits performance bottlenecks.

Learning Objectives

Learn how new access patterns for large content repositories allow for optimizations in file system design.
Understand the importance of providing reliability and scalability for both data and metadata.
Learn about optimistic concurrency on an underlying dispersed storage substrate, and how allows effective metadata management without complex distributed transaction systems.
Learn how this technique allows for distributed writes..
Learn how dispersal allows file system design to be simplified by eliminating the complexity of replication management.