with the default netty settings
storm.messaging.netty.client_worker_threads =1 and 
storm.messaging.netty.server_worker_threads=1
stopology.max.spout.pending=600
i found all spout have many byte[] object in jvm and the complete latency is 
very high in spout---about 12000ms, the whole throughput is very low, so i  
guess there is a bottleneck between workers.

then i increase the thread both of  netty client and netty server,   
storm.messaging.netty.client_worker_threads =20 and 
storm.messaging.netty.server_worker_threads=20,
the whole throughput incease 4* then before, and the complete latency is also 
decrease to 4000ms.

in my topology, 40 spouts( there have 40 partitions), 6 workers, 1 acker.

is there netty is a bottleneck?  and how to set netty threads?



Reply via email to