First put into the shared memory is slower than subsequent one about unidist HOT 1 CLOSED

The root cause of this issue is, before monitor loop has to start, the monitor process has to perform the allocation of shared_memory . In the given example (with times 2.53 and 0.06) almost 2.5 seconds is spend in _allocate_shared_memory() function for the monitor process. Once the allocation is complete after a certain seconds(2.5 in our case) the monitor loop becomes ready to receive unidist operations. Once the monitor loop is up and running all the further puts with similar objects would take similar times.

The initial allocation of shared memory with MPI.Win.Allocate_shared is not time taking. However in _allocate_shared_memory(self): function we are setting -1 to the service buffer because 0 is a valid value. This operation is a time consuming operation. I tried removing this line from unidist source and could see the time taken for the first put would decrease.

Have to figure out if there is a faster way to set the service buffer to all -1, or if this could be removed if this is not necessary.

from unidist.