You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi Genial,
I believe you are correct about the order in which operations get executed. They are reversed simply to make the task of keeping track of them simpler. But this is not incorrect behavior because there is no guarantee of order between concurrent delegate ops. Any time you want to ensure ordering between two delegates, you must have some synchronization. So blocking delegates in the same task are ordered. One could order these delegates as you suggested, but in general you still wouldn't have a guarantee that concurrent delegates would maintain their original order over the network.
Hope that helps.
Hello,
I am confused about your intra-node MessageList design which use
compare_and_swap
to insert a outbound message. (https://github.com/uwsampa/grappa/blob/master/system/RDMAAggregator.hpp#L672)So, the newer message will on the top of the older message in your messagelist like:
and I think we should reverse the MessageList before we process it since we should make sure that the older message should be handled first.
However, I did't find something like that, and I just see you use
compare_and_swap
to grab a messagelist and process it one by one. (https://github.com/uwsampa/grappa/blob/master/system/RDMAAggregator.hpp#L546)The text was updated successfully, but these errors were encountered: