I noted disparate results among the different base file size Mongo benchmarks, so I wanted to see what would happen in a real scenario (at least for us) where the Linux VFS and cache techniques can improve enormously the global performance of the system.

In the case of the kernel compilation, Ext2 has a very low performance for copying files, for the other tests ReiserFS with notail option is the winner but the times are very close to Ext2 and XFS, the difference is less than 2% for make bzImage.

In the last case, where I mixed lseek, read, write and fsync on files of different sizes, the winner is XFS but for a very small difference, less that 8% compared to ReiserFS.

Analysing all benchmarks, it seems that for the common cases, ReiserFS and XFS have a better performance that Ext2 and with the added value of a journaled file system.

What can I say? If you are a home user or own a small server and a relatively fast CPU, use ReiserFS or XFS, both were very stable in our tests and the differences are almost inexistent.