Conference bulova::decw_jan-89_to_nov-90

3067.0. "Client Load for X Session" by DPDMAI::STEINER () Thu Jul 12 1990 09:39

Posted also in the notesfile below:
    
             <<< VAXWRK::$1$DUS6:[NOTES$LIBRARY]VMSNOTES.NOTE;1 >>>
             -< VAX/VMS and more!  *** FOR INTERNAL USE ONLY ***. >-
================================================================================
Note 1984.0          X-Terminal/VAXstation Load on Client???          No replies
DPDMAI::STEINER                                      34 lines  11-JUL-1990 16:53
--------------------------------------------------------------------------------
    Is anyone in the VMS Performance and Test Systems Engineering working
    on general guidelines for Client Load Characterization for X-Terminals
    and Workstations (User Guidelines)?
    
    We have read the internal report by John Buford and some articles in DR
    magazine and are involved in a raging customer debate over cost per
    seat in an DECwindows environment.  The question is...  "What is a more
    cost effective solution to a DECwindows application in image display
    from a DECwindows client to multiple DECwindow servers?"  Is it a large
    Client with X-Terminals, or is it a smaller Client with low end
    VAXstations.
    
    The VUP and memory requirements on a client are substantial!  And if
    very little interaction from the user is required at the display end,
    is the cost per seat for usage of the client resources more expensive
    with a 6000 class client than on low end VAXstations?
    
    As observed in a user test environment, the CPU requirements on display
    startup are between 18 and 24 percent of a VS3100 per window.  Each
    window requires a new process on the client and if many servers are
    active, a bottleneck results, with corresponding delays on the
    X-Session display.
    
    Can we expect a display server process on the X-Client for multiple
    head display on the X-Server?  And would we be correct to assume a
    reduction in load on the Client Processor and by how much?
    
    I realise that it is difficult to estimate the requirements
    generically, without applications specific data, but this is exactly
    what customers want for budgeting cost of systems.
    
    Any experience of others in the field?, or in Engineering?
    
    Phil

    It has been a while since I worked in this area, but since you
    asked, here goes...
    
    To the best of my knowlege, no one in the VMS Engineering space is
    characterizing X-Terminals or the load which they put on the remote
    systems (VMS, Ultrix, etc.).  
    
    I cannot say whether the X-Terminal development team is doing this 
    sort of analysis. (I can't even remember the name of the organization
    because it has changed once or twice since I did the report that you
    read.)  
    
    Unfortunately, I doubt that there is an easy answer to your question
    today.  General guidelines depend on being able to characterize the
    type of work that the user does.  You mentioned that you read my
    report; please study Appendix A *carefully*, particularly the
    description of the applications and operations.  If your customer's
    users do different tasks (I guarantee that they do) then the resource
    usage figures will differ.  By how much?  Nobody knows...
    
    Here is a wonderful opportunity for someone to do Digital and its
    customers are tremendous service: go to a representative sample of our
    DECwindows customer sites and collect data on user work habits (which
    apps are running, where are the apps running, which apps are being used 
    actively, how often does the user switch from actively using one app to 
    another, how often does the user start and stop apps, does this pattern
    change with the time-of-day?) and system resource usage data (what is
    the CPU utilization, how much memory is being used, is the memory
    shared or local, how many users are actually sharing sharable pages,
    how much I/O is being performed, what does that I/O represent e.g. X
    protocol, remote page faults, remote locking protocol, application
    specific I/O?).
    
    Until we can characterize how users actually use DECwindows, lab
    generated "guidelines" are just a shot in the dark...

While not as all encompassing as one would like, there are two papers out
of the LES Systems Engineering System Characterization Group you can look
at to get some feeling for host load.

WINDY>dir oldtmr::secg$reports:vt*.ps

Directory OLDTMR::DISK8:[SECG.REPORTS]

VT1000_REPORT.PS;1  VT1000_WHITE_PAPER.PS;1