In the elasticsearch.yml file in 1.x.x series there used to be a recommendation for setting index.number_of_shards to 1 for single-node clusters. The setting and the recommendation is not there any more in 2.1.0. Is there a reason for that?
In my experience with 1.x.x, multiple shards per index in single node cluster had performance overheads (both speed and memory). But it appears that, at least speed-wise, in 2.1.0 that is not the case. Having one shard per index is significantly slower than 1.x.x. But falling back to default (5 shards per index) speeds it up significantly.