ilkoch008 / cpp_exam Goto Github PK

Here you can see simple wrappers for std::unordered_map, which are supposed to be thread-safe. In methods, that can modify data, added mutex lock in one case, and test_and_set loop in another.

You can see measurement results below:

Mutex with RANGE = 100

atomic_flag with the same RANGE:

RANGE is the value that represents max range for key values (from 0 to RANGE). This parameter has strong correlation with hash-table size, but in this task it has almost no influence on elapsed time (but it would in case of using something like compare_and_swap, I just don't know how to implement in an appropriate way something like this for unordered_map).

n_writers is the value, that represents number of threads, performing modifying operations on map. There is also always one thread, that performs find.

Also worth pointing that in this tables total amount of work has a linear dependence on n_writers, because each thread does the same amount of work. So actual plateau should look like linear increasing.