Conference rusure::math

1235.0. "Anybody think they know how to model an LAVC?" by OLDTMR::STCLAIR () Tue May 08 1990 09:20

I would like suggestions on how you would proceed  to  develop  models

and use them to predict the response of workstations and LAVCs.



I believe the problems can be described by three types of models:



     A.  The mechanical devices  can  be  represented  by  traditional

         models  based  on Newton's Laws of Motion.  For example a the

         positioning of a disk's heads obeys these laws.



     B.  The activity of the silicon chips and software is rule  based

         and deterministic.



     C.  The activity of users can be described by stochastic  models,

         game and queuing theory.





If you agree that the forgoing is true then; all you  have  to  do  is

program  the  first part in FORTRAN, the second in LISP, the last in C

and run the model.  %^)



Is there a language or better still a package (like SIMULA) that would

permit modeling all three areas at once?



I fear there are at least three pitfalls in my logic.  The first  fear

is  that  my  premise is fundamentally flawed.  The second is that the

premise is correct and that the best tools are the ones suggested  but

models  cannot  be  constructed  before  Von Neuman engines are passe.

Finally if I can overcome my first fears that the model is so  complex

and  slow  that  once  started  it  will not finish running before the

design of the particular computer being modeled is obsolete.



If you know of any complete, or even incomplete, models please let  me

know.   If  you have experience with modeling any of these areas would

you please contact me I am sure I can benefit by your experience.

There is a package that is used by a simulation that I stole from
the Ozix performance group called CSIM that may be what you are looking
for.  It layers a number of discrete event queueing simulation primitives
on top of C, so you have the full capabilities of C to mix and
match stochastic and deterministic behaviors.

The simulation program I am using is called IOBUDGET and can be gotten
from Ozix performance.  Try contacting Mike Peterson @ DECWET::MPeterson.

There are some manuals that describe the CSIM package.  Try contacting
RDVAX::MCC and ask for MCC technical report numbers:

	ACT - ST - 252 - 87, rev. 14, CSIM reference manual
	ACT - 126 - 90, CSIM user's guide

CSIM is used in a few other places in the corporation; I think Richie
Lary and some of the folks in Colorado also use it for disk subsystem
simulation.

There are some simple written examples of how the primitives are used spread
around in some of the conference proceedings written by the originator,
Herb Schwetman:

	Proceeding of the 1986 Winter Simulation Conference
		"CSIM: A C-Based, Process-oriented Simulation Language"

	?? conference proceedings
		"Using CSIM to model complex systems"

You might want to look into it; I can't say if this is what you need.

    
    LAVC modeling can be done by VPA (VAX Performance Advisor), and its
    modeling component VME (VAXcluster modeling environment). 
    
    See notesfiles VINO::VPA and HPSRAD::VME for more information.