Revision as of 09:31, 10 July 2007

Distributed Filesystem for Wireless Mesh Networks

Advisor: Anatolij Zubow

Expected Submission: December 2007

Which problem do you want to solve?
- There is no filesystem only based on DHT for decentralized use, like our BRN.
To which domain of Computer Science does this problem relate to?
Why is the problem interesting? Why is it difficult?
- A filesystem is a basic service for many applications on top. It can be used for class- and update-distribution, backup- or storage-service and many more.

At the end of the project, what will you have delivered?
- A basic distributed (high-performance) filesystem, using only the DHT.
- An interface to that filesystem in fuse or nfs, to mount it.
What does it mean that the project was successful?
- The filesystem's performance is sufficient for class- or upgrade-distribution.

How will you approach the problem?
- I'll build the filesystem from bottom to top:
  - abstraction to fuse, nfs or similar
  - fs-layer: c# library, with an ioctl-like interface
  - block-layer: abstract the dht to simple blocks/extends variable in size
  - dht-layer: use the DHT
- All the BRN-magic is done in the DHT, so i'll try to build a clever block structure with some caching to improve everything's speed.
On which contributions of others do you depend on?
- Distributed Hash Table by Anselm Ringleben

@@ Line 42: / Line 42: @@
 ** Implement the Blocklayer.
 ** Implement Direcorys:
+*** openDir
 *** mkdir
+*** rmdir
 *** ls
-*** chdir
+*** <s>chdir</s>
-*** pwd
+*** <s>pwd</s>
+** Implement Files
+*** openFile
+*** unlink
+*** read
+*** write
+*** truncate
+** Implement Fuse Frontend
 * What You plan to do next?
-** Implement Files.
+** fix bugs
-** Build an abstraction to fuse or nfs.
+** push performance to its limits