I measured embedded and reference collections with and without shards. Even without replication, sharding these datasets into 2 shards almost doubles storage space. What is relationship of sharding with increasing storage size?
I tested 2 shards and storage size almost thrice. Multiple shards?