The read to write ratio is often an important design consideration for MongoDB modeling. When reading data, we want to avoid scatter and gather situations, where we have to hit several shards with random I/O requests to get the data our application needs.

When writing data, on the other hand, we want to spread out writes to as many servers as possible, to avoid overloading any single one of them. These goals appear to be conflicting on the surface but they can be combined once we know our access patterns, coupled with application design considerations, like using a replica set to read from secondary nodes.