Following the works of SERVER-119347 and SERVER-119153, we should create a benchmark to have a comparison between the unordered and ordered semaphore implementation so we can assess the real life strenght and weaknesses of each implementation. The idea is to create a workload similar to the mutex benchmark where we have several threads that hold the lock for a variable time, and maybe another one that queues request and see how the code performs. Part of the work for this ticket is to define valuable scenarios for the benchmark, and the other part is to implement them.