Paper #: Week 4 Paper 2
Title: Serverless Network File Systems

(1) Problems
This paper deals with the traditional problem of client server systems by proposing a serverless network file system. The client server model has the inherent problem that it has basically one point of failure - the server. Also, the server becomes a performance and reliability bottleneck. Serverless network file systems try to do away with this problem by balancing the load and the file system functions between the peers (connected to each other with a fast local area network). The authors present such a serverless network file system with their xFS prototype in this paper.

(2) New Idea and Strengths
i) The main idea is to tackle the problem of a one point failure system - the client server system. The authors present a nice brief desciption of why this is a problem and then give a solution to this problem with their xFS prototype.
ii) The xFS prototype is built with the help of the LFS, RAID and Zebra systems. This is a very nice design as it reuses tried and tested systems instead of reinventing from the beginning.
iii) The three principles - distributed control processing, usage of RAID and global file, that the serverless design uses are very well thought out.
iv) The authors actually build a system (though only a prototype as of now) to prove their claims. 

(3) Weaknesses and Extensions
i) The connection of the peers needs to be on a fast LAN. This might not be possible in all settings and hence is a restriction for the system though the authors also propose a NFS solution for this problem.
ii) Security is a problem in the system. The peers trust one anothers kernel to enforce security. This design is not scalable in a heterogeneous non restrictive environment.
iii) The xFS system has not been fully implemented yet. The paper and the results are based only on a prototype. This is a weakness and it would be interesting to check the results of an actual system implementation.
-- END --