Design and Analysis of a Dynamic Extent-Based Allocation Technique for Multimedia File Systems Used in CDN Proxies
Stanislav Sokolov · NORA - Norwegian Open Research Archives · 2006
This thesis investigates the efficiency of extent-based allocator design to satisfy file allocation requests in a CDN proxy cache. The allocator is based on the method inspired by the memory allocators, where free space is managed in chunks of varying size, or extents. The design is tested in a simulation, where a trace of allocation and deallocation events from a content server was submitted to the allocator. Content serves of this type experience high demands on throughput, so their file system must store files in the most efficient way possible. The bottleneck for content retrieval often lies on data transfer rates of the hard disks used in the server. To facilitate fastest possible transfer of a file, it must be read sequentially, in one operation. At the same time, given the large quantity of file, which are present on such servers, space wastage due to incomplete utilisation of large allocation units is not desirable. Our allocator design tries to achieve both, to a certain extent, mutually exclusive goals. The design was implemented and the results we obtained in the course of simulation, show that we managed to achieve these goals, creating an allocator that displays properties, favourable for contiguous file placement, while keeping space wastage at its minimum. Additionally, the allocator is memory-efficient and has small bookkeeping and computational overhead. The use of our allocator in a CDN proxy file system will allow to keep the data transfer throughput at maximum speed, while utilising the storage space in an efficient manner. The reader of this thesis will learn about the allocator semantics and have a detailed introduction into a specific allocation algorithm, QuickFit. This study outlines several venues of improvement and opens for a further empirical study of complete system, based on the allocator presented in this thesis.