From:	INTRSL::WAMPLER      "Kurt Wampler"  8-JAN-1988 12:14
To:	ARISIA::EVERHART
Subj:	Follow-up on VMS 4.6 AUTOGEN problem

Follow-up message on VMS 4.6 AUTOGEN problem with page & swap files:

Last November, I sent out a message to system managers on the GE network
about a problem we encountered with VMS 4.6 AUTOGEN.  Since that time, I
have worked with TSC in Colorado Springs to reproduce the problem on their
system, and talked to George Claborne, the author of AUTOGEN, at the DECUS
Symposium in Anaheim last month.  They have acknowledged the problem, and
generated an internal SPR for it.

Here's a brief description of what we found wrong:

The 4.6 AUTOGEN attempts to inspect all current page and swap files by doing
a $ SHOW MEMORY /FILES and temporarily redirecting SYS$OUTPUT to a scratch
file to capture this information.  It then opens and reads each line in the
scratch file, trying to locate the page & swap files, their names, and sizes.
Unfortunately, they didn't do a $ SHOW MEM /FILES /FULL (this would have
explicitly told them which files were paging versus swapping).  To distinguish
between pagefiles and swapfiles, they look for the presence of the string
"PAGEFILE" in the filespec.  There are 2 problems with this simplisitic scheme: 

  1) If the site has not named all its page and swapfiles after the PAGEFILE.XXX
     and SWAPFILE.XXX naming convention, it will correctly distinguish between
     page & swapfiles.

  2) If the site has the string "PAGEFILE" somewhere else in the filespec (i.e.
     the device name or the directory name), it can fail to correctly
     distinguish between page & swapfiles.

In our case, we had located our secondary paging & swapping in a directory on
another disk called DRC0:[PAGEFILES].  So when AUTOGEN parsed the filespec
"DRC0:[PAGEFILES]SWAPFILE.002" it thought it was another pagefile instead of
a swapfile.  That's where all our trouble came from.

The developer admitted that the existing code was a real kludge, and indicated
that he was working to get an additional lexical function (or option to
F$GETSYI) that would allow him to retrieve the desired information in a more
efficient and robust fashion.

There is some additional logic in that section of code that attempts to
"balance" the distribution of paging and swapping space that looks suspicious
to me as well.  The documentation is not clear as to when this balancing logic
is activated.  For the time being, I'm sticking with the PAGEFILE=0 and
SWAPFILE=0 options in MODPARAMS.DAT that disable the page/swapfile adjustment
features of AUTOGEN.

Kurt Wampler
Mgr. CAD Software Development @ GE/Intersil
(408) 996-5426 or 8*823-5426