Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Serverless Network File Systems (xFS)

Paper Review

By Syed Jibranuddin (50026775) In this paper authors proposed serverless distributed filesystem which provided filesystem services by utilizing the the computers cooperating as peers, and believing that the centralised distributed filesystems have bottleneck., as it is expensive and all the read/write misses goes through the central server. In order to provide better performance and scalability than these traditional file systems like AFS, NFS, netware, Sprite, the authors implemented a prototype serverless network filesystem called as xFS. There were 3 major motivations behind implementing xFS, first was the opportunity provided by fast switched Ethernet/LANs, the second one was the expanding demands of users and the last one was fundamental limitations of central server systems. XFS relied on previous works as scalable cache consistency, cooperative caching , disk striping(Raid and Zebra) and log structured filesystems. Key Features: The idea of taking advantage of the cooperative caching to server client memory seems pretty interesting. Use of RAID, by adding a feature of grouping the storage servers into groups on the features provided by ZEBRA. Performance of the system was improved without upgrading the hardware by the use of map and their distribution over the areas. Striping of files over group of servers instead of striping it over all servers in the system improving the availability as each group maintain its own parity. It improved the old fashion server in term of performance, scalability, and availability. Limitations : Measurements were quite limited as the workloads werent the real ones, they are just benchmarks which performs better than real workloads. Lacked some couple of features like - Recovery was not completed, lack of distributed cleaner, and it doesn't calculate or distribute new manager or stripe group maps. Wasnt able to identify and solve the security issues. Deficient information regarding the results of testing and performance.

the size of the log segment wasnt pretty well discussed. Discussion is Class : In terms of the implementation of xFS use of RAID5 (i.e. distributed parity ) in place of RAID4 would not have affected the performance of xFS. First Major difference between ZEBRA and xFS is that xFS is serverless where as ZEBRA used Servers and the second major difference is ZEBRA used RAID but was not scalable as striping of files were done all over the disks whereas in xFS stripe groups were formed. Performance of IMAP in locating a file seems to be providing overhead but the inclusion of Imap vs the bottleneck on the system leads to conclusion that imap added to perfomance of system. xFS out-shined the old server system like NFS and AFS in terms of performance, scalability, and availability and the cons associated with the centralised servers were removed and cost was also less as compared to older systems.

All in all it was a nice, well organised and interesting paper...

You might also like