Hacker News new | comments | show | ask | jobs | submit login
A Study of Linux File System Evolution (usenix.org)
12 points by marbu on Mar 24, 2013 | hide | past | web | favorite | 1 comment

I am blown away by this study! Kudos for all involved, it must have been a huge effort.

After having read it one time and reread parts a second time, here is my take-away:

1.) When having to chose a file system I would go with Ext4: (a) It is probably the most used Linux file-system, (b) it is 2nd in number of patches which is a very good sign, while it has half the code size of XFS (which is the 1st in number of patches).

2.) When dimensioning a file system (a) be generous (space, number of inodes) (b) use reliable hardware in order to NOT run into bugs that will be provoked by lack of resources.

Here are some facts and quotes from the study that marked me most:

1.) "The patch percentages are relatively stable over time; newer file systems (e.g., Btrfs) deviate occasionally; bug patches do not diminish despite stability."

2.) "A high fraction of bugs occur due to improper behavior in the presence of failures or errors across all file systems; memory-related errors are particularly common along these rarely-executed code paths; a quarter of semantic bugs are found on failure paths."

3.) "Concurrency bugs are responsible for nearly all deadlocks (almost by definition) and a fair percentage of corruptions and hangs."

4.) XFS development seems to be quite active (the study goes just to 2011). How comes that the amount of XFS patches is so high? Higher than for Ext4 and BTRFS? Is this FS used so much or is this driven by companies?

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact