Calculated CPU time of C++ is different from actual time to solve MILP model via Gurobi

Question

I am solving a MILP model in C++ using Gurobi 7.5.2. More specifically, I am determining the CPU time of my C++ program via the following commands:

• clock_t start_time;
• double elapsed_time;
• start_time= clock();
• model.optimize();
• elapsed_time = (double) (clock() - start_time)/CLOCKS_PER_SEC;

When I compare the elapsed_time with the Gurobi runtime, the Gurobi runtime does show the actual duration (measured by a chronometer) but the calculated elapsed_time of C++ is way higher. Why is the elapsed_time not the same as the Gurobi runtime and the actual duration (measured by the chronometer)?

Answer 1

This is more of a C++ question, so you should rather look for similar questions on stackoverflow.

A common issue is that clock() is not accurate for short durations, so you may want one of the high resolution C++ timers instead, as shown in an answer here.

EDIT: as commented by user3680510 and Stradivari, this is due to the use of multithreading, as clock() measures the CPU time. The C++ timer above (or C's time()) will behave correctly.

Answer 2

clock() measures CPU time instead of real time. In order to measure real time in C++ use:

auto start = std::chrono::steady_clock::now();

//do stuff

auto end = std::chrono::steady_clock::now();

auto milliseconds = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();