The evaluated CSP, N-Queens, exhibits a high node throughput since processing a single store is a fast operation. As a consequence, the accesses to and maintenance of the worker’s pool play an important role on scalability. On one hand the
overhead incurred from releasing work on the pool is large, preventing better efficiency. On the other hand, more time is spent retrieving stores from the pool, including stealing operations which happen at a higher rate. Hence, there is a high rate of failed steals (local and remote). Nonetheless and after the exposure of such bottlenecks, it was possible to reduce the overhead due to the release operations by simply increasing the interval at which they should happen.