Paper #21 Serverless Network File Systems Problems This paper disucesses the design and implementation of a serverless network file systems(NFS). Traditional NFSs are centralized and have a number of disadavantages such as performance bottleneck, reliability bottleneck due to single point of failure as well as expensive. On the other hand, a serverless file system allows us to get rid of all these problems. Strengths & New Ideas 1. The paper makes a strong argument for the need for a serverless network file system. It describes the risks and problems associated with a traditional centralized network file system. It shows that in the future such centralized systems will not scale well, will be unreliable, inefficient and expensive. 2. The paper describes in great detail the design of a serverless, distributed network file system. It lists the required components of such a system and how they correspond to a tradition centralized system. The paper also describes how existing technologies such as RAID, LFS, Zebra and Consistent Hashing can be used to implement an efficient serverless NFS. 3. The paper discusses xFS, the first prototype of a serverless NFS implemented by the authors. They discuss the various data structures of xFS and give a detailed description of how xFS handles a request for a file. The description explains helps to understand the internal working of a serverless NFS very well. 4. They also discuss the current limitations of each of the components. The discussion of these limitations (expensive log cleanups etc) helps the reader to identify possible research areas if they are new to the field or beware of pitfalls if they are trying to implement their own serverless NFS. Weaknesses 1. Lack of Security The serverless approach is only appropriate in restricted, controlled environment as there are a lot of kernel calls between the collaborating peers to improve performance. Kernel calls are a major security issue and in an open environment would result in an highly insecure system. 2. Require Fast Network The machines require fast networks. Highly pervasive ethernet networks do not provide the features neccessary(high bandwidth, multiple transmissions) to make a serverless NFS efficient. Once again, this results in the serverless NFS being restricted to smaller networks which have extremely fast networks. 3. Not a fully, functional prototype (not tested on industrial strength application) The prototype does not have some crucial functionality such as crash recovery procedures that are neccessary for any NFS. Besides, it has not been tested to serve commercial applications. They can only be validated as a promising research prototype.