RE>As an intersting sidenote, have you seen much difference between HPFS386
RE>with 2MBs cache or i.e. 4MBs.  I haven't seen much using IOSTONE and
RE>several other benchmarks?  I wonder why as particularly IOSTONE should
RE>take advantage of the additional cache.

None of them will stress 4MB. Look at IOSTONE's source code. It should have
been in the archive.

 396 files taking 983040 bytes total.

 File Size  Num Files
 =========  =========
 256        128
 512        64
 1024       64
 2048       64
 4096       32
 8192       32
 16384      8
 32768      2
 65536      2

Albert Shan did a lot of testing on the various benchmarks about a year
ago using SPM/2. I went behind him and verified a lot of it. IOSTONE turned
out to be rather pathetic.

If HPFS386 uses the HPFS 128k allocation on open rule that was listed,
I think in the redbooks, all the files fit in memory on open.  If it also
follows the 30 percent write cache dirty rule 2MB would be just as good as
4MB for the purposes of the test.  The bottom line is that only your normal
program mix will tell you what you really need for cache.  I can't see IBM
being blackmailed on this HPFS issue much longer.  There are too many
reasons to replace it with a different file system; especially for the
enterprise users.  Therefore there is probably not much reason for someone
to write a test to prove the theories.  A couple of associates and I were
toying with the idea in our diminishing spare time and decided it was not
worth it.

--Lynn

 * SLMR 2.1a * Closed captioned in hex for programmers where available.

--- DB 1.39/004485