|  | 
   You're not the only one that is confused -- I suspect the wording on
   the tables should be "minimum" and "default", I don't know where the
   "dynamic" is coming from here.  I further suspect this confusion may
   have been due to a cut-and-paste from a SYSGEN table that once included
   the settings of some PQL SYSGEN parameters...  The PQL SYSGEN parameters
   are paired, with both a "PQL_D" default and a "PQL_M" minimum form.)
   When no PQL parameter is specified on a process creation, the default
   value is used.  And the lowest setting one can have is the minimum
   value.
   Table 1-6 lists the minimum amount of *contiguous* global section
   pages, unlike the previous tables that indicate total requirements.
   The page tables used for global sections, just like disk sectors,
   can become fragmented.  The only way to clear this is to deinstall
   and/or delete some resident stuff, or (unlike disks) to reboot...
   Set the minimum values as required in the documentation, ignore the
   default (or "dynamic" settings) remembering when making the changes
   to use SYS$SYSTEM:MODPARAMS.DAT and AUTOGEN, and using the ADD_ or
   the MIN_ syntax on the entries -- do not, for instance set GBLSECTIONS
   to 10, use ADD_GBLSECTIONS=10 or MIN_GBLSECTIONS=currentvalue+10.  Also
   remember to account for other usage of the parameter -- GBLPAGES and
   GBLSECTIONS parameters are pretty cheap, and it does not normally
   hurt typical OpenVMS operations nor memory usage to overconfigure
   these settings by 10% to 20%, or more.  (And when in doubt, "add
   more memory".  See the performance manual for the specific overhead
   involved with increasing these settings...) 
   (I'd recommend whomever is in control of the KITINSTAL add a couple
   of f$getsyi("FREE_GBLPAGES") and f$getsyi("CONTIG_GBLPAGES") calls,
   to verify these settings.  (The underlying requirement for these
   checks has, however, likely been entirely superceded by the INSTALL
   changes made in the OpenVMS V6.2 release -- INSTALL now tests for
   room *before* deinstalling the old image.  And this change makes it
   rather harder to end up with no DCLTABLES installed when sufficient
   contiguous pages do not exist.  Of course, the C++ kit likely also
   installs on pre-V6.2 systems, too.)
   I'd log a QAR against the C installation documentation.  This is
   rather confusing...
 |