Adaptive Metadata Management and Flexible Consistency in a Distributed In-memory File-System

Abstract

Distributed file-systems are a popular storage abstraction for cloud-computing applications. They provide generic data access for different applications in order to pass information between computing nodes and to save computation results persistently. The performance of distributed applications depends on data-consistency protocols and meta-data management, but these factors of influence are often statically configured in distributed file-systems. In this paper, we describe EFS, an in-memory file-system that manages meta-data and consistency by flexibly adapting to file-access patterns.