File System Supporting Arbitrarily Sized Allo ation

The majority of today’s filesystems use a fixed block size, defined when the filesys-tem is created. If an allocation request is not a multiple of the block size, space is wasted when parts of blocks are left unused. This thesis looks into the viability of a filesystem with arbitrarily sized blocks, called grains. The file system uses an existing allocator, implemented by Stanislav Sokolov, based on the Quick Fit main memory allocation algorithm. To limit the scope of the thesis, the file system is implemented using FUSE, and lacks support for growing and fragmented files, directory hierar-chies and crash recovery. However, even without these featurees, the file sytem is still useful for a few scenarios, where file sizes are known in advance and the size distribution is favorable for the allocator. Examples of this are web- and mail servers. The thesis also runs some simulations based on a recorded mailserver workload, and compares results to the same workload on Ext2. 4 A knowledgements I would like to thank my primary supervisor Željko Vrba for his guidance and pa