Read contention problem with an exclusive lock

Sometimes it is a nightmare for a C++ developer to handle multiple threads in an application. If the number of threads is in thousands, synchronization is a nightmare.

As a C++ developer, we use a mutex to synchronize and after some efforts, we succeed to so synchronize these threads with a mutex.

Now comes a performance issue. One of those I will discuss here.

Read Contention:

There is a problem with exclusive lock is read contention

If we have multiple reader threads trying to access a shared resource(let's assume Database), in case of the exclusive lock each reader thread will take a lock on resource and process and release the lock. Meanwhile, other reader threads will wait for the resource to be free to get a lock and so on.

So let's assume each reader thread is locking resource for 1 sec and we have 1000 or 10000 reader threads. You can calculate total time required to an application to complete just reading from the database.

The solution in C++:

In C++14, there is a new class introduced as shared_timed_mutex and in C++17, shared_mutex.

How to simulate this and solve:

I simulated this performance problem with 1 writer thread and 2 reader thread. And found a significant difference in total run time.

For more details, check out the link in the first comment.