On Tue, 2005-12-13 at 21:43 +0100, Andrzej Bialecki wrote: > > Most of the time we deal with very large files, with sequential > access. > Only in few places we deal with a lot of small files (e.g. indexing). > So, I think the best would be an FS optimized for efficient > sequential > write/read of large files.
But beware what happens if you run more than one task per machine. Each individual task might be sequential but several in parallel will generate plenty of disk head movement that approximates parallel IO -- especially on a filesystem that uses small blocks and a driver with poor read-ahead support. -- Rod Taylor <[EMAIL PROTECTED]>
