UNIX for Advanced & Expert Users

I would be taking my masters degree project next year. So I am now in the process of thinking something that could help me get more in depth knowledge and will be a useful one in practice.

I have thought about creating shared file system for clustered environments (like GPFS,GFS).

I have below doubts before taking this idea any further.
1. Is it feasible to complete it in one year if I start some home work now ?
2. Is there a way to reuse any code from linux ? (the one from GFS)
3. Any documents to understand the implementation of any existing shared file systems .
4. Best book to start for file system implementations.
Please post your thoughts. That would guide me to choose the better option.

Thanks.

Well, it is hardly worthwhile unless you have some extension or innovation to apply. After you dissect the open source of a SOTA SFS, you may find some ways to make it lower latency or higher bandwidth. Perhaps, using cluster-friendly networking like broadcast UDP. You could add in compression, perhaps somewhat like zip subtrees. You could make it a distributed peer file system aggregating a heterogenous ix of disks on all hosts of the cluster, with dynamic mirroring, where extra copies are made if a host disappears, and multiple copies exist for low churn high use files, or hybrid mirrroring, where one side of the mirror is compressed and the other takes the query load. Of course, a year is sufficient, depending on you anf your schedule.

Thanks for you expert view. Though it would be hardly possible to match the performance of the commercial products, at least this would give me the understanding i am looking for. And innovation has to be found only after understanding the thing.

Can you suggest any book that could help me understand the file system implementation or any material about this topic.

I'd get one running and look at the open source. Book writing is big business, but code and playing tells you more. Also, there are many good pieces on open source products online. Get started first, and buy the book if you find you need help.

I also fancied a UFS variant where directories are trimmed, maybe even organized for hash or tree search. Another idea is a dir that can hold small inodes, files and directories within them, recursively, with dynamic expansion and initial/utility compression. Saves both inodes and pages. All invisible to old software, give or take a funny inode #.

Compression like zip is the thing that really fascinates me, as zip seeks to the compressed file you want, and as CPUs are faster than disks, the flow rates through zip may exceed most disk speeds. You could even chop a file into super-pages and compress each separately, for easier update and seek time within the file. Integration with the VM would allow all the decompressed file or superblock to be cached to VM. It could be rewritten back to disk as a redundant uncompressed copy in case of more query.

Thanks for you tips here. I'll first start before getting deep into to optimizations :-)

Thanks again for your valuable tips.