As we saw in threaded applications that we coded in 213, the need for locks results from possible race conditions on the values of variables. In this case, the race condition is on the value of diff. Note that locks are not especially beneficial to the efficiency of parallel programs because they serialize the execution around them, creating a possible bottleneck while one thread executes the code in the lock and others simply spin while waiting for the lock to free up.
This comment was marked helpful 0 times.
rbcarlso
It's a good thing that while cores aren't getting faster, we're still getting more space. The more different cores/threads can work in their own space and save combining for the very end, the less negative impact locks will have on performance.
As we saw in threaded applications that we coded in 213, the need for locks results from possible race conditions on the values of variables. In this case, the race condition is on the value of
diff. Note that locks are not especially beneficial to the efficiency of parallel programs because they serialize the execution around them, creating a possible bottleneck while one thread executes the code in the lock and others simply spin while waiting for the lock to free up.This comment was marked helpful 0 times.
It's a good thing that while cores aren't getting faster, we're still getting more space. The more different cores/threads can work in their own space and save combining for the very end, the less negative impact locks will have on performance.
And now a joke:
"Knock-knock," "Who's th--," "Race condition!"
This comment was marked helpful 1 times.