The AFS default of 1 file for 10KB of cache space is based on an old
study showing that the average file size is 11KB, and leaving a little
margin. If you have reason to believe that your average file size is
much larger than is typical, you can adjust the number of files which
AFS will use by specifying the -files switch on the afsd command line.
If you set this value too low, you will not completely utilize the
cache space. For example, supposing your cache is 64000 KB and you
use a chunk size of 64 KB, if you specify -files < 1000, you will
never fully utilize your cache. Even if you specify files > 1000, you
may not fully utilize the cache, depending on your average chunk size.
Specifying -files somewhat larger than absolutely neccessary (say,
double) should have neglible performance impact. If it's an order of
magnitude larger than neccessary, I would recommend reducing it.
You may have read UNIX adminstration treatises which discourage very
large directories. In this case, those recommendations do not apply,
since the cache manager does not access the chunks via the standard
directory lookup, but accesses the inodes directly. Also note that
lookup in large AFS directories is often faster than lookup in
large UFS directories because the directory structure is different.
On the other hand, "ls -l" of a very large AFS directory may be
unpleasant.
Lyle Transarc 707 Grant Street
412 338 4400 The Gulf Tower Pittsburgh 15219
