NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/lang/ada/2391

Text File | 1992-08-20 | 3.7 KB | 68 lines

Newsgroups: comp.lang.ada
Path: sparky!uunet!cis.ohio-state.edu!udecc.engr.udayton.edu!blackbird.afit.af.mil!mawhelan
From: [e-mail] (Michael A. Whelan)
Subject: Dynamic Task Creation/Control
Message-ID: <[e-mail]>
Originator: mawhelan@scgraph
Sender: [e-mail]
Nntp-Posting-Host: scgraph.afit.af.mil
Organization: Air Force Institute of Technology
Date: Thu, 20 Aug 1992 17:48:57 GMT
Lines: 55

Wisdom Sought!

    I am currently working on my MS here at AFIT.  The title of my thesis
is "A Real-Time Intelligent System Architecture Implemented in Ada".  The
'direction' of the thesis is to provide a way to implement what are termed
'Real-Time AI Sytems' in Ada.  Generally, real-time AI systems only
address the issue of agenda management, and not deadline guarentees.  I
am trying to add the ability for the 'Intelligent' part to control, the
'real-time' part.
    Please do not ask for an definition of an 'intelligent' agent, I am not
attempting to start a philosphical debate.  Also, the 'concept' of a
real-time intelligent system is not new and I can provide references to
anyone who asks.
    What I am currently attempting to do is allow for the dynamic creation,
execution, scheduling, prioritzing, etc. of both periodic and
non-periodic tasks.  I am assuming (well maybe not assuming but using an
educated speculation) that most work done in an 'intelligent' system is
done by 'procedural' processes/tasks.  The goal of the 'intelligent' part
then is to figure out what 'procedural' part to execute and be able to
have it done in real-time, meaning guaranteed deadlines.
    Periodic tasks are scheduled using a varity of 'scheduling policies' but
the underlying 'scheduling theory' is rate monotonic.  Because, I am
assuming that I do not know a-priori what the periodic task set looks like,
(i.e. how many tasks, the periods or each, etc.) and because there is a
limited number of priorites available, ( Verdix on a Sun gives me 100 ),
the actual assigned priority depends on a scheduling policy.  Policies as
they exist now are static, normal distribution, and linear.  Static as the
name implies assumes a predined non-dynamic priority assignment.  Normal
distribution tries to achieve what the name implies; the closer to the mean
period, the more tasks will have the same priority.  Linear tries to assign
the fewest tasks to the highest priority and the most tasks to the lowest
priority.  Non-periodic tasks are scheduled using some techniques for imprecise
algorithms and since this is long enough already, I'll provide references if
requested.
    The area where wisdom is needed deals with the dynamic creation of tasks.
Since Ada tasking essential keeps the memory assigned to an access type
task until termination of the task that has the access type defined, I am
forced to create a 'task-control-block' that keeps track of which tasks
are currently being used.  In addition, I have to try and reuse these
tasks for different 'procedural code'.  Any suggestions?
    Also, when 'it is decided' to terminate a task ( done normally because it
has passed it's deadline ), I need to reclaim the task-control-block and
be able to reuse it.  However, I am at a loss to come up with a good way to
insure it is in a known state.  Again, any suggestions?
    I know this short note has left a lot of unanswered questions, but I
hope it is sufficent to give those of you wiser than I some idea of the
problem domain.  If you have any thoughts you would like to share on the
idea of dynamic task creation and control please do so.  It has been my
experience that half the solution to any problem is asking the right people
for ideas.


Capt Mike Whelan

e-mail: [e-mail]