Re: Suggestion for review: a taxonomy of thread safety

From: Nick Landsberg (SPAMhukolauTRAP_at_SPAMworldnetTRAP.att.net)
Date: 09/24/04

  • Next message: Drebin: "Re: Where can I find a design pattern for paging through a list?" 
    Date: Fri, 24 Sep 2004 12:52:29 GMT

    Jim Rogers wrote:
    > Nick Landsberg <SPAMhukolauTRAP@SPAMworldnetTRAP.att.net> wrote in
    > news:7EK4d.625843$Gx4.47450@bgtnsc04-news.ops.worldnet.att.net:
    >
    >
    >>To possibly turn this discussion to somewhat differnt
    >>aspect:
    >>
    >>"If multi-threading is the answer, what's the question?"
    >>
    >>In my experience, threading is useful if:
    >>you have threads (flows of control) which must wait
    >>on slow devices, e.g. disks, networks, etc.
    >>(Not the only case, the most obvious case.)
    >>
    >>In this situation, you may have some thread perform
    >>useful computations while the another thread is waiting
    >>for the response.
    >
    >
    > I have successfully implemented event-based robotic control
    > systems using multiple threads of control. My program had no
    > problems with locking and unlocking of data. The reason my
    > program had no such problems was because I used a language
    > that implicitly takes care of those issues.
    >
    > The language is Ada. The multi-tasking implementation is
    > very robust, very flexible, and highly reliable.
    >
    > Examples:
    >
    > On a robotic system some sensor information, such as
    > vehicle heading, is delivered at a constant rate which
    > may exceed the system's sampling needs. In that case,
    > the system needs to read the most recent value from
    > the sensor, and does not need to be flooded with values
    > it cannot handle.
    >
    > One task can read from the sensor at the sensor's
    > specified rate. It will write to a shared memory area.
    > Another task will read the shared memory area at a
    > slower rate; the rate which the system can process the
    > data. How does one set up a shared memory area such that
    > memory writes are locked, but memory reads are not (allowing
    > reads to be as fast as possible)? One Ada solution is:
    >

    [SNIP]

    What you posted, Jim, has many things going
    for it. In addition to the language support
    (the more I hear about ADA the more I'm impressed),
    what you described was a very disciplined design
    to begin with. (Having the language support looks to
    have made the implementation easier.)

     From a language independent point of view, the
    design is more critical. Take a design, for
    instance, where there are multiple writers
    and multiple readers of multiple shared
    resources. Whenever I run across
    one of those, I start to worry, for a number
    of reasons. It might be legitimate, but often
    it is due to some newbie deciding to multi-thread
    his/her existing code without thinking through
    the ramifications. In the extreme, you can wind
    up with slower execution than if you had left
    it single-threaded in the first place.

    Thanks for the post.

    NPL 

    -- 
"It is impossible to make anything foolproof
because fools are so ingenious"
  - A. Bloch
  • Next message: Drebin: "Re: Where can I find a design pattern for paging through a list?"

    Relevant Pages

    • Is it possible in a DataGrid?
      ... I come from another language where a control called a ... a more sophisticated datagrid could accept multiple win ... In effect you had the ability to set each control above ...
      (microsoft.public.dotnet.framework.windowsforms.controls)
    • Re: Three Dog Problems
      ... If you can control the meaning of words, ... This is too narrow a view for the role of language. ... Orwell was just.... ... His idea of history and its control, culture and its control, ...
      (sci.cognitive)
    • Re: definition of a highlevel language?
      ... to the language of the problem domain. ... while control: ... And if you would propose this solution to a control system engineer, ... So an adequate high level language, ...
      (comp.lang.python)
    • Re: Increasing efficiency in C
      ... > You keep control of everything. ... C is a 'lightweight', *flexible* language. ... Depends upon the machine, but yes, imo one should ... nor 'plug-board' programming ...
      (comp.lang.c)
    • Re: Transfering control without growing control context -- i.e. how to implement GOTO
      ... the control context is normally referred to as the "control ... language doesn't support Proper Tail Recursion. ... In JavaScript the control stack will grow, ... Globally declared named functions become properties of the javascript global object so they can be referenced, but still with property accessors, as members of that object. ...
      (comp.lang.javascript)
    Loading