Notes.txt

1) By using selective signalling we can reduce the cpu usage at the sender side and imporve response time
of the coflict-free calls such as deposit and register
Here are the information:
    https://www.rdmamojo.com/2014/06/30/working-unsignaled-completions/
    https://www.rdmamojo.com/2013/06/08/tips-and-tricks-to-optimize-your-rdma-code/#Work_with_selective_signaling_in_the_Send_Queue

2) Make sure the synchronization logic for the state manipulation in courseware is correct
Right now, I only lock students relation for checking the permissibility of enrolls and when registering student.
    Can we synthesize this given the conflict graph? (dependency edges between connected components and free vertices create synchronization)

*3) Sometimes, there are missing received writes (myabe due to the completion queue overflow??)

4) Poor performance for dependent methods is due to dependency calculation NOT waiting.

5) Almost .7 microseconds speedup in response time when doing polling with 1000 batch for bank and courseware.
However, some writes are missing!

6) For multi synch groups make sure to set memory to more that 12GB per node.
6) For multi synch groups make sure to increase the time in the script.