...
Now we have to figure out when to release the blocking partition.
Of each task manager one task will read the subpartition and send a release message to the subpartition when finished. The other tasks which doesn't read just sent a "I am done" message to the subpartition. We know the number of tasks, so we know when every task is finished.
This approach currently does not work for native iterations (see https://issues.apache.org/jira/browse/FLINK-1713).
In the case of a simple Map().withBroadcastSet() we achieve the following results on the IBM-cluster:
suite | name | average_time (ms) |
broadcast.ibm-power-1 | broadcast.01 | 6433 |
broadcast.ibm-power-1 | broadcast.02 | 6162.66 |
broadcast.ibm-power-1 | broadcast.04 | 6459 |
broadcast.ibm-power-1 | broadcast.08 | 6635.66 |
broadcast.ibm-power-1 | broadcast.16 | 7002.66 |
Compatibility, Deprecation, and Migration Plan
...