Tunefs -a 16 doesn't work right

[Larry Blair]

In Sun-Spots v7n303 I wrote about some testing that I did for
optimization.  Thanks to Art Hays <lsr-vax!art at uunet.uu.net>, I found that
the results that I published were incomplete and that there is a bug in
BSD 4.2 and 4.3 (including all levels of SunOS) that prevents changing the
maxcontig parameter of the filesystem ("tunefs -a 16") from working
properly.

While changing maxcontig will eliminate the insertion of rotational delays
in the indirect blocks of a large file, it does not cause the blocks
directly accessed through the inode to be contiguous.  The use of "tunefs
-d 0", which is recommended by the controller manufacturers, does properly
remove the rotational delays.

My performance tests were performed with a few, very large, files.  The
ratio of direct to indirect blocks was low enough to not affect the times
appreciably.  Retesting with many 90K files (the largest file that doesn't
use indirect blocks) shows the problem.  Reading through an od of the
filesystem reveals the bug.

Larry Blair   ames!vsi1!lmb   lmb at vicom.com