Major Description
Looks like we have to omit per-thread buffer for batches that are passed to data streamer.
Large per-thread buffer size may have negative affect on latency in some cases (small batch or small per-node buffer size).
Attachments
Issue Links
- links to
