|
Sounds like a good counterpart for http://jira.mongodb.org/browse/SERVER-1594 Those two would allow do make drivers less complex (routing moves entirely into mongos, also for replica sets and distributed reads) and if mongods were tagged all kinds of nifty things could be done based on that foundation.
Maybe each mongod could generate some "performance" tag based on bogmips, RAM, disk i/o, etc. and let the mongos know so the cluster would somehow know which machine is the "biggest/fastest" machine(s) in the cluster. Further down the road it might be feasible to semi-automatically spit out some suggestion for how to (re)configure the cluster for optimal performance.
|