Previous | Next --- Slide 19 of 29
Back to Lecture Thumbnails
Mayank

I think there are certain details missing here. Does the following sound correct?
When the lock is created, the l->status array should be initialized such that l->head = 0, l->status[0] = 0 and all other elements are 1.
During unlock, we should also set l->status[my_element] = 1, because my_element is a circular increment. If we do not set l->status[my_element] to 1, then after the first round of circular increment, all the elements in l->status will be 0.

kayvonf

@Mayank: thanks. I'll add the detail to the slides.

tpassaro

I'm not sure how this could be better than a ticket lock implementation. In high contention areas, you would assume in a ticket lock that cache values would be updated as long as more contention for the lock was coming in. Since threads will need to wait no matter what, there can be cache coherency done in the background to hide these this latency. This, and the fact you need to allocate some room for an array seems like a waste when ticket locks work so simply.