Global ordering is key, as briefly mentioned. To me, the proven distributed hash table systems like Bigtable, Spanner, HBase, RocksDB have it, and those that do

Global ordering is key, as briefly mentioned. To me, the proven distributed hash table systems like Bigtable, Spanner, HBase, RocksDB have it, and those that don't have it, aren't as strongly consistent (Riak, Cassandra, Mongo, etc..) Having global ordering lets you reason about where your data needs to go, where it needs to be, and overlay different access patterns to see what will work. Heavy write load? Have more shards. Heavy read load? replicate each shard more. Random read/write? Nothing will get you as close to uniform load distribution as global ordering.