AtomicDFS is a Marie-Curie funded project under the action Intra-European Fellowships for Career Development. As such, the goal of the project is twofold: (a) Conduct innovative research in the areas of Distributed Computing and Networks towards the implementation of efficient Atomic Distributed File systems, and (b) Provide the foundation for the fellow to follow a successful career, through a training program he will undergo in the host institution.
The ultimate goal of the project is to investigate the existence of efficient, highly consistent distributed file systems in asynchronous environments where fail-prone processes communicate via exchanging messages. A distributed file system suggests the replication of stored data (files) to geographically dispersed storage devices. Replication ensures both survivability and availability of data while avoiding potential performance bottlenecks. With replication comes the challenge of maintaining consistency when the data are altered or read concurrently by multiple operations. Atomicity guarantees to provide the illusion of a single sequential storage space as if it was accessed on a local machine. Whenever a file is read we expect to obtain: (a) the changes of the last preceding write/append operation, and (b) a copy of the file that is as recent as the one obtained by the last preceding read operation. To reach this goal the project will explore two different aspects of a file system: (i) File Management including file fragmentation and file distribution among the replicas, and (ii) File Consistency including the file retrievals and atomic guarantees. For all directions we plan to obtain theoretical bounds on the performance of computational approaches to solve this problem and develop algorithms that meet our proposed bounds. Next, we plan to implement and test the proposed algorithms in both single processor simulation environments, as well as in real-time network platforms. Such developments will contribute towards the production of a practical file system on top of cheap commodity hardware with strict, provable guarantees. The outcomes of the project are expected to be reported in technical publications including manuscripts in well established, international conferences and journals in the area of distributed computing. Such publications will expose the fellow in the scientific community and they are expected to attract the attention of the industrial world.
The training plan of the fellow can be divided in three pillars:
Through the project the fellow will have the chance to participate in training activities that will enhance his skills in all three areas.
