Originally posted by: Trino

Hello, I have an algorithm in c++ that calls cplex to solve a certain formulation and at the end of each iteration it may add some additional constraints, then call cplex again, and  so on...

I'm limiting the number of threads in 1 by using the following code:

cplex.setParam(IloCplex::Threads, 1);

At the end of each iteration I call the method getTime() of the instance of IloCplex that I use to solve my model in order to accumulate the total time. What happens is that sometimes the total time calculated using this method is very different from the one I calculate by myself using the library -ltr of linux (sometimes double the time!). What I though is that cplex may be using another thread, but I checked and I was setting the number of threads, so I don't know why it is giving me these results, can somebody give me some insight about this?

Thanks.

Could it be that you're using different clock types with the two timers?  By default, cplex will use wall-clock time for the clocktype parameter.

Sorry for my ignorance, but what is the "library -ltr of linux" that you refer to?

Originally posted by: Trino

This is the Posix RealtimeExtension and it allows me to use the following functions to get the time http://pubs.opengroup.org/onlinepubs/007908799/xsh/clock_gettime.html

In cplex I'm using the default option which gives me the wall time, while in this library I use the CLOCK_MONOTONIC (more precise than the wall-clock). But could it be the difference being the double??

Just to be sure: you are aware that IloCplex::getTime() only returns a timestamp, not the elapsed time? So your code should look like this:

double const start = cplex.getTime();
cplex.solve();
double const elapsed = cplex.getTime() - start;

When you observe that CPLEX reports twice as much elapsed time as CLOCK_MONOTONIC, what is the order of magnitude of the elapsed times? Are we talking second, milliseconds, nanoseconds here?

Does the problem persist if you use CLOCK_REALTIME instead? What happens if you use gettimeofday() to time CPLEX execution?

Originally posted by: Trino

uhm...I was not aware of that, I was considering that it returns the elapsed time. So that should be it!

I'll will test and give a fideback here.

Thanks!