Correct me if I am wrong... I believe the dirty bit needs to be turned off?
And, if it is, should it be done at step 2?
This comment was marked helpful 0 times.
kayvonf
@raphaelk: You are correct! The dirty bit should be off at the end of this transaction.
EDIT: The slides are fixed (you may have to refresh the page)
This comment was marked helpful 0 times.
GG
Question: In the previous method, the owner should send dir revision to the home node and then the home node mark the request node as present in the corresponding dir entry. But here, the home node mark the request node as present right after step 1 and the dir revision msg disappears.
Why we need an additional dir revision msg in the previous methods but not need in this one?
This comment was marked helpful 0 times.
lazyplus
@GG: I think this method just assumed that anyone who had requested the memory would eventually become one sharer. And thus the home node can update directory after step 1.
I think the previous method could be improved in the same way. There is no need to send the dir revision message in the last step.
Anyway, the previous method is just a naive but working approach. This is why we have this slide talking about some optimization. We should not assume that the previous method is optimized already.
Correct me if I am wrong... I believe the dirty bit needs to be turned off? And, if it is, should it be done at step 2?
This comment was marked helpful 0 times.
@raphaelk: You are correct! The dirty bit should be off at the end of this transaction.
EDIT: The slides are fixed (you may have to refresh the page)
This comment was marked helpful 0 times.
Question: In the previous method, the owner should send dir revision to the home node and then the home node mark the request node as present in the corresponding dir entry. But here, the home node mark the request node as present right after step 1 and the dir revision msg disappears.
Why we need an additional dir revision msg in the previous methods but not need in this one?
This comment was marked helpful 0 times.
@GG: I think this method just assumed that anyone who had requested the memory would eventually become one sharer. And thus the home node can update directory after step 1.
I think the previous method could be improved in the same way. There is no need to send the dir revision message in the last step.
Anyway, the previous method is just a naive but working approach. This is why we have this slide talking about some optimization. We should not assume that the previous method is optimized already.
This comment was marked helpful 0 times.