I was watching a presentation on high performance computing and starting at 24:55 one of the presenter states that synchronization by acquiring a lock requires a context switch down to the kernel of the OS (not even the one of the virtual machine, if the application runs in one). I would appreciate a brief explanation of why that is the case and why it cannot be handled inside the JVM (in case of Java apps).
Asked
Active
Viewed 130 times
1 Answers
0
From a brief look at the video around 24:55 it appears they are discussing the "acquire mutex-enter critical section-release mutex".
Synchronisation relies on the fact that acquiring the mutex is an atomic operation. As pre-emption occurs in the kernel, it is necessary for the kernel to handle the mutex so as to guarantee its atomicity when multiple processes attempt to acquire/release the mutex.
See here for more information: Are mutexes really slower?
