Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!shelby!agate!ucbvax!f4.n494.z5.fidonet.ORG!011HFAA.WITSVMA
From: 011HFAA.WITSVMA@f4.n494.z5.fidonet.ORG (011HFAA WITSVMA)
Newsgroups: comp.sys.transputer
Subject: (none)
Message-ID: <37.27B64BD2@puddle.fidonet.org>
Date: 11 Feb 91 14:19:25 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 118

To :- Transputer Users
 
Processor Utilisation - An Alternative Approach
 
"Inside The Transputer", D.A.P Mitchell et al.
Blackwell Scientific Publications, Oxford, 1990
ISBN  0-632-01689-2
 
Mitchell et al. describe an alternative routine for estimating processor
utilisation. The fundamental difference between this approach and Andy
Rabagliati's SUPERVISOR procedure is in the implementation of the idle
counter process. Mitchell's version of the process, IDLETIME, is listed
below :
 
-------------------------------------------------------------------------
 
.. SC IdleTime
.. COMMENT
This PROC extracted from :
"Inside The Transputer", D. Mitchell, et al.,
Blackwell Scientific, Oxford, 1990, p. 77
 
PROC IdleTime (INT IdleCount, ExtraCount, Semaphore)
 
  INT L.IdleCount, L.ExtraCount :
  INT Back.Low.Ptr, FrontLowPtr :
 
  GUY
 
    LDC 0              -- initialise counters
    STL L.IdleCount
    LDC 0
    STL L.ExtraCount
 
    :REPEAT
    LDL Semaphore      -- read Semaphore
    CJ  .END           -- if zero terminate
    LDLP FrontLowPtr   -- read low priority queue registers
    SAVEL
    LDL FrontLowPtr    -- and compare front one with MinInt
    MINT
    DIFF
    CJ  .INCIC         -- if queue empty inc idle count
    LDL L.ExtraCount   -- otherwise inc extra count
    ADC 1
    STL L.ExtraCount
    J  .PASS
 
    :INCIC
    LDL L.IdleCount    -- inc idle count
    ADC 1
    STL L.IdleCount
 
    :PASS              -- add to end of low priority
    LDC 2              -- active queue
    LDLP 0
    STARTP
    STOPP              -- deschedule this process
    J  .REPEAT
 
    :END
    LDL L.IdleCount    -- return counts and finish
    STL IdleCount
    LDL L.ExtraCount
    STL ExtraCount
:
 
-------------------------------------------------------------------------
 
This procedure is used in very much the same way as the idle counter
process is used by the SUPERVISOR. As shown below, IDLETIME is launched
in parallel with the rest of the processes on the Transputer. On
terminating (by setting SEMAPHORE to 0), the procedure returns the value
of the idle counter (IDLECOUNT), and also an indication of the time
taken up by the execution of the procedure itself (EXTRACOUNT).
 
-------------------------------------------------------------------------
 
.. Initialize
Semaphore   := 1
IdleCount   := 0
ExtraCount  := 0
 
PAR
 
  SEQ
 
    ... Processes to be evaluated
 
    Semaphore := 0 -- terminate IdleTime ()
 
  IdleTime (IdleCount, ExtraCount, Semaphore)
 
-------------------------------------------------------------------------
 
As for the SUPERVISOR process, it is necessary to establish a calibration
value relative to which the actual IDLECOUNT value can be measured.
 
I have made use of the SUPERVISOR procedure (well, the one that appeared
on this mailing list in October 1990) and the approach mentioned above.
For a selection of examples designed to force a wide range of processor
utilisation, the results, in both cases, were **identical**. So take your
choice ...
 
P.S. I would recommend Mitchell et al. to anyone who is interested in
programming in assembly language.
 
Regards
 
Craig Faasen
 
011hfaa.witsvma@f4.n494.z5.fidonet.org
 
"And if I live forever, I won't ever be the same
 For I have seen the sun rise in Zimbabwe ... "
--  
uucp: uunet!m2xenix!puddle!5!494!4!011HFAA.WITSVMA
Internet: 011HFAA.WITSVMA@f4.n494.z5.fidonet.org