Thread Priority

[LN]

I am running a process that is sending data trough an interface but it seems
that windows NT interrupts
the communication, messing up the spacing between data blocks. Is there anything
I can change in my
software that can stop windows NT from interrupting when data is being transfered?

[Andy Friedman]

Not sure I understand the question. Are you saying that the timing of
receiving your data is so sensitive, that normal CPU rotation caused by the
OS thread scheduler is screwing up your data? Please explain.

Andy

"LN" <lnarvaez@west.raytheon.com> wrote in message
news:3a251005$1@news.devx.com...
>
> I am running a process that is sending data trough an interface but it
seems
> that windows NT interrupts
> the communication, messing up the spacing between data blocks. Is there
anything
> I can change in my
> software that can stop windows NT from interrupting when data is being
transfered?

[LN]

I am sending the data and there has to be a fixed amount of time between data
blocks. The data is
coming out of a file. I can send many blocks of data without any problem
but every once in a while a
data block will have to be sent again because the timing was not right and
the receiving system would
not return a good response to my application. We were able to see the data
on a logical analyzer when
this error happened and the spacing between the blocks was sometimes more
than twice what the
receiving system was expecting. We think that the OS is interrupting in the
middle of a transmition
delaying our timing whish is just a couple of milliseconds.

"Andy Friedman" <devxnews@zr1.com> wrote:
>Not sure I understand the question. Are you saying that the timing of
>receiving your data is so sensitive, that normal CPU rotation caused by
the
>OS thread scheduler is screwing up your data? Please explain.
>
>Andy
>
>"LN" <lnarvaez@west.raytheon.com> wrote in message
>news:3a251005$1@news.devx.com...
>>
>> I am running a process that is sending data trough an interface but it
>seems
>> that windows NT interrupts
>> the communication, messing up the spacing between data blocks. Is there
>anything
>> I can change in my
>> software that can stop windows NT from interrupting when data is being
>transfered?
>
>

[chris]

well, the short answer is yes.

immediately prior to sending data, bump up the priority with setpriorityclass
and setthreadpriority. the highest values being REALTIME_PRIORITY_CLASS/THREAD_PRIORITY_TIME_CRITICAL
which would cause your thread to preempt even os tasks such as flushing disk
caches etc.

this is what you are asking - to have your thread not be interrupted by ANYTHING
- even the os.

if you do this however, be very careful. if you remain in this high priority
state too long, nothing will respond - mouse, keyboard, anything. as a result,
i would certainly wrap up such a priority shift within an RAI to ensure its
reset to a lower class/priority at the appropriate time.

now, that said, i would highly question such a design move. implementing
a data protocol dependent on such finely grained timing (a couple of ms)
on a non-hard real time os is not the best move (imho).

"LN" <lnarvaez@west.raytheon.com> wrote:
>
>I am sending the data and there has to be a fixed amount of time between
data
>blocks. The data is
>coming out of a file. I can send many blocks of data without any problem
>but every once in a while a
>data block will have to be sent again because the timing was not right and
>the receiving system would
>not return a good response to my application. We were able to see the data
>on a logical analyzer when
>this error happened and the spacing between the blocks was sometimes more
>than twice what the
>receiving system was expecting. We think that the OS is interrupting in
the
>middle of a transmition
>delaying our timing whish is just a couple of milliseconds.
>
>"Andy Friedman" <devxnews@zr1.com> wrote:
>>Not sure I understand the question. Are you saying that the timing of
>>receiving your data is so sensitive, that normal CPU rotation caused by
>the
>>OS thread scheduler is screwing up your data? Please explain.
>>
>>Andy
>>
>>"LN" <lnarvaez@west.raytheon.com> wrote in message
>>news:3a251005$1@news.devx.com...
>>>
>>> I am running a process that is sending data trough an interface but it
>>seems
>>> that windows NT interrupts
>>> the communication, messing up the spacing between data blocks. Is there
>>anything
>>> I can change in my
>>> software that can stop windows NT from interrupting when data is being
>>transfered?
>>
>>
>

[Andy Friedman]

>
> now, that said, i would highly question such a design move. implementing
> a data protocol dependent on such finely grained timing (a couple of ms)
> on a non-hard real time os is not the best move (imho).
>

My thoughts exactly.

Andy

[Danny Kalev]

Andy Friedman wrote:
>
> >
> > now, that said, i would highly question such a design move. implementing
> > a data protocol dependent on such finely grained timing (a couple of ms)
> > on a non-hard real time os is not the best move (imho).
> >
>
> My thoughts exactly.

I second that too:-)

Danny

[Jim Howard]

You might consider porting your application to run under a real time operating
system. I have implemented time critical applications under DOS using some
hacks, but I'd never try it under windows. Your only hope is to boost your
thread priority to the max and hope that windows doesn't crash as a result.

You might look at getting a single board computer to plug into a host pc.
Your NT box could control the overall operation, and the single board computer
could do the time-critical communications.

If you have to analyse your output on a scope then you shouldn't be using
Windows. Windows is a great OS, but it's not an RTOS.

Jim