Monitor Contention

Monitor contention is the primary cause of lack of scalability in multi-processor systems. Monitor contention is typically caused by multiple threads holding global locks too frequently or too long. To detect these scenarios, the profiler may enable the following three types of event notifications:

• A thread waiting to enter a monitor that is already owned by another thread issues a MONITOR_CONTENDED_ENTER event. This event indicates possible performance bottlenecks caused by frequently-contended monitors.
• After a thread finishes waiting to enter a monitor and acquires the monitor, it issues a MONITOR_CONTENDED_ENTERED event. This event indicates the amount of elapsed time the current thread has been blocked before it enters the monitor.
• When a thread exits a monitor, and discovers that another thread is waiting to enter the same monitor, the current thread issues a MONITOR_CONTENDED_EXIT event. This event indicates possible performance problems caused by the current thread holding the monitor for too long.

In all these three cases, overhead of issuing the event is negligible compared to the performance impact of the blocked monitor operation. The profiler agent can obtain the stack trace of the current thread and thus attribute the monitor contention events to the parts of the program responsible for issuing the monitor operations.