[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference vaxaxp::vmsnotes

Title:VAX and Alpha VMS
Notice:This is a new VMSnotes, please read note 2.1
Moderator:VAXAXP::BERNARDO
Created:Wed Jan 22 1997
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:703
Total number of notes:3722

531.0. "Q on working set size and page faults" by HGRD01::HOMANSANG (Keith Ho, ISE HK @HGO) Fri Apr 25 1997 07:26

    I have a questions on the working set size:
    
    Am I correct that if the WSDEF and WSQUOTA are increased (both smaller
    than the WSEXTENT), the number of page faults of a process could be
    reduced provided that the system has free memory (as showen in SHOW
    MEMORY)?
    
    I tried to increase a process WSDEF and WSQUOTA and re-login.  The peak
    working set size was still the same and page faults were not reduced
    (shown by SHOW PROCESS/ACC).  I don't know why the system could not
    give more phsical memory to that process (as there are quite a lot of
    free memory) so that the page faults can be redcued.
    
    (BTW, the SYSGEN parameter WSMAX is large enough.)
    
    If I'm not correct, then exactly what control the page fault related to
    the working set size?
    
    Thank you,
      -- Keith.
T.RTitleUserPersonal
Name		DateLines
531.1My first cut at an explanation.CPEEDY::FLEURYFri Apr 25 1997 09:4925
    RE: .0
    
    While the workingset parameters themselves dictate the number of faults
    to a degree, there are other factors as well...
    
    1) WSDEF is the "size of the box" that the process starts with.  This
    	is the number of possible pages that can be in the working set to
    	begin with.  (Note that pages are faulted in one(cluster?) at a time)
    
    2) If the number of pages in the working set = WSDEF, then the "box"
    	must grow.  This is done in WSINC increments.
    
    3) The workingset grows at AWSTIME which is usually a multiple of
    	QUANTUM.  This is important to remember as a lot of people change
    	QUANTUM to a smaller number without allowing the workingsets to
    	grow more rapidly, thus possibly incurring MORE pagefaults.
    
    Overall, if the total number of pagefaults for the running program is
    reaches a stable number, or they are of theh demand zero type, there is
    little that can be done to eliminate them.  Pagefaults are not
    necessarily bad.  Paging to a disk or excessive pagefaults are not
    good.
    
    Dan
531.2WSDEF = size of the working set LIST, not the working set itselfFRSTSC::TLAUER"I've been designed multi-asking."Tue Apr 29 1997 03:5016
When a new process is created (i.e. via login), the WSDEF does NOT specify the
number of valid pages this process should start it's life. WSDEF just
initializes the working set LIST size, that is, it prepares some data structures
to be able to hold as much as WSDEF pages lateron. Filling a process's working
set list with pointers to valid pages (WSLEs) is done via the normal
page-faulting mechanism during image activation etc. So, regardless what you set
WSDEF to, if you re-login to see the effect, VMS will have to do exactly the
same amount of page mapping to create your process environment and to activate
some images.

If you really want to examine a process's working set dynamics, you have to
eliminate these initial steps producing unavoidable paging from your
measurements.


-- Thilo
531.3in other wordsGIDDAY::GILLINGSa crucible of informative mistakesTue Apr 29 1997 20:189
  Keith,
    In other words, if you run the same process with increased WSDEFAULT and/or
  WSQUOTA and see no difference in the number of pagefaults, then you can 
  conclude that the pagefault rate is NOT due to the working set being too
  small. There will be some finite number of unavoidable pagefaults required
  populate the woking set in order to to execute a program. If the working
  set is large enough, once it's filled the pagefault rate should slow down
  or stop.
						John Gillings, Sydney CSC