Instead of using hand-over locking, we can observe the paths that the two transactions (in this case, two writes, one to node 3 and one to node 4) take and determine whether or not the operations need to be serialized.
One possible solution is to first check whether the paths of the transactions running in parallel conflict or not. If they do, then we serialize the operations. Else, we can run them in parallel. Note that the observations can be done in parallel, since they aren't changing the data structure (only determine which operations are reads/writes). It's a small overhead, but it can allow the operations to run in parallel much more frequently.
Instead of using hand-over locking, we can observe the paths that the two transactions (in this case, two writes, one to node 3 and one to node 4) take and determine whether or not the operations need to be serialized.
One possible solution is to first check whether the paths of the transactions running in parallel conflict or not. If they do, then we serialize the operations. Else, we can run them in parallel. Note that the observations can be done in parallel, since they aren't changing the data structure (only determine which operations are reads/writes). It's a small overhead, but it can allow the operations to run in parallel much more frequently.
This comment was marked helpful 1 times.