From: ahemavathy (ahemavathy_at_hotmail.com)
Date: Sun Feb 22 2004 - 13:27:55 PST
This paper is about a new file system called the Log Structured file system. The goal of this file system is to improve the read and write disk accesses which is becoming a bottleneck to overall system improvement inspite of tremendous improvement in processor speed and memory. The data is stored as log on the disk : the only factor which will affect its efficiency is .... how to main large free space.
LFS uses a i-node map data structure where it maintains the location of each i-node. Free spaces are maintained by copying live data in a compacted form. Each segment has a summary block which has data which tells the system if the segment is live or not. Thus this system eliminates the need for a free list. The segment cleaning is started whenever the no. of clean segments falls below a threshold. They further make it efficient by incorporating a policy based on cost and benefit. Crash Recovery is by checkpoints and roll-forward recovery methods.
The authors feel that existing file system structures do not benefit from improvement in processor or memory while the log file system will improve. I liked the paper as it was easy to read. But I don't agree with the authors that just by improving the disk writes this file system is better than the others. I would have been more impressed if the reads were improved. Because anyway the writes are cached in the file caches and writes can be postponed until the processor is free. Most of the existing file systems also have the crash recovery techniques like checkpoints for which they used a log structured system for temporary storage. This should be more than enough to recover from disk crashes.
This archive was generated by hypermail 2.1.6 : Sun Feb 22 2004 - 13:26:44 PST