We are using MapReduce. Latency will depend on your cluster and the query; it's just a regular MapReduce operation from Hadoop's point of view.