Looks like it uses only one core?
Will you show the code, which generates the workload? Does your question involves any of Couchbase products?
We have a thrift server with two methods:
Add and Get
Add add some []int64{} to bucket with key.
Get get nth element from bucket with given key.
The pic shows is our prod server(16 cores). It seems that the core are not used efficiently.
We read the source code and add ?kv_pool_size=16 as couchbase cluster addr suffix. Is this correct?