Is there any suggested ratio for dividing the number of handler threads between normal RPC and service RPC (HDFS)?

[#1]

How should we reassign the current number of hdfs.namenode.handler.count to dfs.namenode.service.handler.count and hdfs.namenode.handler.count after activating service RPC?

For instance, we currently having 240 handler threads - should we set dfs.namenode.service.handler.count to 120 and hdfs.namenode.handler.count to 120 or do you suggest other ratios?

[#2 · Accepted solution]

@Arpit Agarwal

Thanks for helping me out. For others' information: ~50% of dfs.namenode.handler.count is a reasonable value for dfs.namenode.service.handler.count.