> Though number of connections in no way reflects the cpu usage of a
> broker;
>
what are you thinking here? I am thinking that typically there will be
a direct correlation between cpu usage and number of connections. In
particular a broker with no connections will use very little cpu, add
a steady producer via a connection and things start to happen, another
connection a consumer or producer and more utilisation will occur etc.
Agreed a plugable strategy is best but on the 80/20 rule I am thinking
raw number of connections is a reasonable metric. Consumers and
producers may already be partitioned by the application. Am I missing
something here?

Reply via email to