Timer object initiating function over extended period of time using too much CPU memory

I am attempting to run a timer object that refers to a function that reads and plots a data point from a pH probe. I plan on running this timer object for up to 3 days, but after 2-3 hours matlab and the computer crashes because CPU usage slowly increases to 100%.

To start the timer I'm using:

h.measuretimer = timer
set(h.measuretimer, 'executionMode', 'fixedRate', 'Period', 4, 'TimerFcn', 'Cori_read')
start(h.measuretimer)

My attempts at fixing this problem have been: 1. Using the "return" function at the end of the code being called upon by the timer object.

Result: CPU crashed after running for 2-3 hours

2. clearing and deleting the timer object after each single measurement, then restarting it.

To do this I wrote: stop(h.measuretimer) delete(h.measuretimer) clear h.measuretimer clearvars return

Result: CPU crashed after running for 2-3 hours

3. Removing the timer object all together and replace with a while loop.

Result: Program ran for 3 days with no problems, CPU usage stayed below 1%. However, this solution prevents further applications I hope to use with this program because the measurements are interrupted when I perform other actions.

My questions are: 1) Why is this happening? 2) Is there a way to properly end a script and remove all memory it used? 3) How do you properly use a timer object over an extended period of time?

Thank you

[Matt Fig]

How long does it take Cori_read to execute? Is Cori_read a script or a function? Why are you using the string instead of @Cori_read?

And where is the code you show above? Hopefully not in Cori_read! I am trying to make sure you are not creating many timers inadvertently.

[Sean]

Cori_read takes less time to execute then the 4 seconds allotted for the timer object. There is an error that comes off the pH probe periodically, that the script checks for by rerunning the check, but this is minimal in time.

Cori_read is a defined function held in its own .m file that is referred to in the timer object code.

I'm not sure what you're referring to in the 3rd questions.

Thanks for the quick response.

[Matt Fig]

Cori_read may take less than 4 seconds, but when using fixedRate you are also counting the queue lag and any other processes. Please give this TimerFcn a try and see if anything suspicious happens....

'disp(''In''),Cori_read,disp(''Out'')'

By the third question, I am asking why you are using a callback string instead of a function handle.

[Sean]

This code is held in its own function that is called on by a button push from a GUI.

 function startMeasurementsCori()
 h.measuretimer = timer
 set(h.measuretimer, 'executionMode', 'fixedRate', 'Period', 4, 'TimerFcn', 'Cori_read')
 start(h.measuretimer)

it is a toggle button with the code

 function MeasurePhCallback(src,evt)

 button_state = get(src,'Value');
 if button_state == get(src,'Max')
 set(src,'String','Stop Measuring')
 startMeasurementsCori

 elseif button_state == get(src,'Min')
 set(src,'String',"Start Measuring')

 end

[Jan]

What exactly does "crash" mean? The CPU usage is going slowly to 100%, but this is not a crash. Do you get an error message or core dump?

I assume there is either a missing pre-allocation in the Cori_read function, or you append to a file, which is opened and closed repeatedly. You can use the profiler to see, which commands need more and more time.

[Sean]

crash in this case is the computer locking up and being non-responsive. The matlab program itself will error out with an abundance of errors referring to java memory, including:

At java.awt.EventDispatchThread.pumpEvents(Unknown Source)

 Exception in thread “AWT-EventQueue-0” java.land.NullPointerException:HDC for component
 At sun.java2d.loops.DrawGlyphList.D

Those were all I was able to gleam out of the non responsive computer.

[Sean de Wolski]

If we could have the code for Cori_read that would be ideal. But yes, otherwise please try a simple timerfcn and see if the CPU still eventually goes to 100.

[Sean]

Within the function for the button toggle that refers to Cori_read, I have defined the variables:

 h.phlog = (1:100000)*NaN;
 h.timelog = (1:100000)*NaN;
 h.timestart = clock; 
 h.number = 100000;

The code for Cori read is:

 function Cori_read()
global h

 %This code opens the pH meter, h.s, retrieves a number, records the time %the number was recorded, and sets it to a variable, h.out
fopen(h.s);
fprintf(h.s,'*IDN?');
h.time=etime(clock,h.timestart)/60
h.out=fscanf(h.s);
fclose(h.s)

 %This is my error check for the number coming out of the pH probe. The %number range I am looking for all have 5 elements to them before str2num, %so this safe guards against any bad numbers coming off the probe

 while numel(h.out)~=5
fopen(h.s);
fprintf(h.s,'*IDN?');
h.time=etime(clock,h.timestart)/60;
h.out=fscanf(h.s);
fclose(h.s);
end

 h.phvalue=str2num(h.out);

 h.phlog(h.number)=h.phvalue
h.timelog(h.number)=h.time

 % plot a limited length of the data points
if h.number<99950
   plot(h.timelog(h.number:h.number+50),h.phlog(h.number:h.number+50)
else
   plot(h.timelog,h.phlog)
end

 h.number=h.number-1
end

[Sean]

h.s is the ph probe. To set up h.s. I use the code:

 h.s = serial('COM4')
set(h.s,'BaudRate',38400,'Terminator','CR')

Upon running the code

 fopen(h.s);
fprintf(h.s,'*IDN?');
h.out=fscanf(h.s);
fclose(h.s)

h.s values are:

Communication Settings

Port: COM4

BaudRate: 38400

Terminator: CR

Communication State

Status: closed RecordStatus: off

Read/Write State

TransferStatus: idle

BytesAvailable: 0

Values Recieved: 5

Values Sent: 6

Answer 1

[Walter Roberson]

Don't use plot() each time through the timer callback. Instead, create a line object the first time through, and after that set() the XData and YData of that line object to update it.

Do you really need to plot all 100,000 points? The graphics effort for that is going to be non-trivial and get worse as you gather more points. At 100,000 points your resolution will be what, 100 points per pixel? Consider plotting only every h.number/1000 pixel (assuming a plot window 1000 pixels wide).

Also, as you go you are having to change the x-axis limits each time because you are plotting more and more time. Consider picking a time-window, last N seconds only, and displaying that, with xlimmode set to manual and you adjusting the xlim from time to time but not every update.

[Malcolm Lidierth]

You could plot then just update the YData values - much faster than a high level plot.

Opening/closing the file each time could be avoided by passing a file handle to the callback and closing it on timer deletion.

Also don't use clock - it's not reliable because the OS may reset system time during the 3 days.

As @Sean asked, does this happen with a trivial (e.g. empty) callback?

Answer 2

[Daniel Shub]

If h.number ever exceeds 100000, and it looks like it might since you do the following check

if h.number<99950

Then you phlog and timelog are not preallocated and growing on every timer callback. This could take a lot of time. Further, plotting 100000 points is a lot. Even if each point is 1x1 pixels, you would be using 20% of your monitor.

[Sean]

So the program is never plotting any more then 50 data points at any one time. That is what the check does. Each data point is put into this array of NaN's which is 100000 points long. NaN's are never graphed so they don't show up. What shows is the real data points put in place of the NaN's. What is graphed is controlled by the ever decreasing h.number and the 50 points in front of it.

h.number never exceeds 100,000. It is set at 100,000 and decreases to zero. When it hits zero, the program will fail out and end, but I hope to stop the program by that point.