(Excerpts from a message dated 09-17-99, Will Honea to Murray Lesser)

Hi Will--

ML>      Deleting the leading "rem " from the line in CONFIG.SYS (leaving
ML> the install-provided: "DISKCACHE=D,LW") and rebooting, I get 3204
ML> KB resident system space and 512 KB of disk cache.  As you will
ML> recognize, the additional 4KB (one page) for resident system is in
ML> the noise, and can be ignored.  This data led me to believe that
ML> there was no default assignment of "DISKCACHE" memory in my system,
ML> in spite of what it said in on-line help!

WH>Last time I had OS2MEMU up, the D parameter was working as advertised
  >- but I also have a couple of FAT partitions.  That may impact the
  >assignment as opposed to floppy-only systems.

    Your post aroused my curiosity, so I ran some more tests.  I have a
small partition (51 MB) that I call my "pseudo virtual drive" since it
gets formatted by STARTUP.CMD.  I use it as a working Drive:  All the
junk left over from compiling (or from running Netscape/2) vanishes with
STARTUP.CMD, saving me a lot of file housekeeping.  So, for these tests,
I formatted it FAT instead of the usual HPFS, thereby giving me one FAT
partition on my hard drive (this system - ThinkPad 365XD - has only one
hard drive).

    When I booted with the DISKCACHE= statement REMmed out of
CONFIG.SYS, I got essentially the same thing as with no FAT drive.
However, when I booted with the "default" DISKCACHE=D,LW, I got an
increase in resident system memory from 3184 to 5568 Kb.  However
OS20MEMU said that memory allocated to disk cache remained at 512 Kb!  I
read this as the system having allocated almost 2400 Kb to the FAT cache
and its supporting driver, but OS20MEMU didn't want to tell me about it!
When I rebooted with DISKCACHE=2048,LW I got resident system memory of
5592 Kb (within the noise of the previous value), but OS20MEMU's report
of memory allocated to disk cache went up to 2560, somewhat verifying my
previous assumption as to where some of that 2400 Kb went to.

    After I had run the above experiments, I reformatted my working
drive HPFS.  Then, I realized that I hadn't covered one case: I didn't
have a floppy drive attached for any of the tests.  My ThinkPad has a
built-in CD-ROM drive, but the floppy is an external drive that must be
plugged in before rebooting, for the system to see it.  At home, the
ThinkPad sits in its port replicator, and is always attached to the
floppy drive, the Zip drive, an external display, keyboard and mouse,
and the printer.  But I am currently on the road, and the external
floppy drive is riding in my carry case.  So I attached it, rebooted,
and ran OS20MEMU with and without the DISKCACHE= line REMmed out of
CONFIG.SYS.  No change in resident memory size nor memory allocated to
disk cache.

    Thus, my tentative conclusions are:  OS/2 (at least at Warp 4 +
FixPak 5) is a lot smarter than I thought it is.  With "D" calling for
the default memory to be allocated to diskcache, the line is ignored if
there aren't any FAT partitions on the hard drive.  When there is one
FAT partition, the system supplies only 2 MB of diskcache, rather than
the 4 MB that the "help" file says my 40 MB of "system memory" should
have generated!  (I don't know if the 4 MB maximum is an on-line
documentation error, or would have been allocated if more than one FAT
partition were on my hard drive.)  If the DISKCACHE= line is REMmed out,
the system doesn't assign any memory to the diskcache, irrespective of
whether or not there are any FAT partitions on the hard drive(s).  I
reiterate that I am using Warp 4 plus FixPak 5; my results may not apply
to other levels of OS/2; YMMV.

    My attitude toward on-line documentation is unchanged:  Too much
information is omitted for it to be useful for much of anything, and
what information is present should be verified by test before drawing
any conclusions from it.  (Note:  My experience to date indicates that
this sad state of on-line documentation is not limited to that furnished
by IBM; I have on-line documentation that came with commercial products
produced by non-IBM sources that are even worse.  One virtue of IBM
desktop software products is that hard-copy documentation can usually be
obtained if one is willing to pay for it.)

    Regards,

        --Murray

 
___
 * MR/2 2.25 #120 * Curiouser and curiouser.

--- Maximus/2 2.02