I read previous posts regarding what does "took" measure. Does it including the time to distribute the query and merger the result?
I run some geo-point search benchmarks for a cluster(3master, 4 data nodes, 1 client, 4cores 16G RAM VMs). The cluster with replica shards seems like show better scalability than the cluster without shards (just allocate the primary shards to 1/2/3/4 nodes) in terms of average "took" value. Is that reasonable?