We’ve found the same thing. A significant amount of cpu time is being taken up by unlocking (?) within the disruptor queue.
From: Rahul Mittal [mailto:rahulmitta...@gmail.com] Sent: 24 September 2014 16:16 To: user@storm.incubator.apache.org Subject: High CPU usage by storm workers Hi We are experiencing some difficulties in bringing down the CPU usage for our storm topologies.We are getting only 4k-5k tuples per second The worker uses around 2 to 3 cpu cores. After we profiled our worker we found out that majority of the CPU time is taken by ReentrantLock.unlock() (around 40-50%). I have attached the profiler snapshots. the configuration we are running with are: no of workers = 1 bolt parallelism = 3 spout parallelism = 3 (kafka Spout) Topology_reciever_buffer_size =8 Topology_transfer_buffer_size =32 Topology_executor_buffer_size =16384 Topology_executor_send_buffer_size =16384 Also we are not facing any problems with throughput . Capcity and execute latecy for each of the tasks seems to be fine. With Regards RAHUL MITTAL
