The Design And Implementation Of Journaling Filesystem On KYLIN

The filesystem is an important part of a modern operating system. It is crucial that the filesystem is reliable and fast, in order to meet today's demands. However it is still possible that computers shut down unexpectedly as the result of hardware or software failures or because of power outages. When this happens, traditional filesystems have to scan through the whole filesystem content, to find inconsistencies. On large partitions this process, of recovering to a consistent state, can take very long and for huge partitions it can take many hours. For many appliances this is not acceptable, so we need a mechanism to speed up this lengthy recovery process after an unclean system shutdown. This mechanism is provided by journaling filesystems which can recover to a consistent state in several seconds.Now, Redhat Enterprise Linux AS with EXT3, Novell SuSE linux with ReiserFS, SGI IRIX with XFS and IBM AIX with JFS, joumaling filesystem has become a very important part of server operating system. However Kylin still does not have a good joumaling filesystem and it is emergent to design and implement a reliable and fast joumaling filesystem on Kylin.Firstly, this paper analyses the features of popular journaling filesystems and the Kylin's kernel frame, especially Kylin's filesystem and virtual filesystem. Secondly, we design and implement KFS joumaling filesystem with an eye to the features of Kylin. In order to improve the directory searching performance of KFS, HTree based indexing directory is designed. In order to layout files on a filesystem contiguously, reservation window based allocator is involved. Finally, we analyse the performace of KFS with three benchmarks: Bonnie++, PostMark and Tiobench .