Hi, *The problem:* A colleague says having 300 workers and 600 executors is a bad idea even if 300 of the 600 are acker bolts. Processing bottleneck, according to him.
*My question: * So far I didn't experience a performance issue, but given the nature of the application, number of tuples, processing and the number of emits can become more demanding. I can understand if a bottleneck is due to my number of bolts being greater than the number of workers. But if my bolts = number of workers, then does creating an equal number of extra ackers really consume so much memory and processing power, that I would have to increase the number of workers to double? *The scenario:* I have a single spout that emits around 20K tuples (I know it should be 1 per spout, but in this scenario it's a necessity). There are 300 bolts receiving the tuples with fields grouping, and 300 ackers. 76 slots are available. On the Storm GUI, there are around 600 executors being shown, which I reckon are my bolts + the acker bolts. >From Nathan Marz' answer <https://groups.google.com/forum/#%21topic/storm-user/uJt-ym5xe60>, setting the number of workers = number of ackers seems a safe bet. Also, he says a bottleneck is application specific <https://groups.google.com/forum/#%21topic/storm-user/kz_p2DITakE>, and that he planned to improve the worker parallelism (wonder if it is implemented yet). -- Regards, Navin