Multithreaded applications take advantage of increasing core counts to achieve high performance. Such programs, however, typically require programmers to reason about data shared among multiple threads. Programmers use synchronization mechanisms such as mutual-exclusion locks to ensure correct updates to shared data in the presence of accesses from multiple threads. Unfortunately, these mechanisms serialize thread accesses to the data and limit scalability.
Often, lock-based programs do not scale because of the long block times caused by serialization, as well as the excessive communication overhead to coordinate synchronization. To reduce the impact on scalability, programmers use fine-grained locking, where instead of using a few locks to protect all shared data (coarse granularity), they use many locks to protect data at a finer granularity. This is a complex and error-prone process11,12 that also often impacts single-thread performance.
No entries found