Investigating performance of RocksDB Cloud as a KV store

[jarohen]

On the theme of the separation of storage and compute (likely in combination with extensive use of write-through disk caches).

(from @refset) other options worth thinking about are:

1. OpenAsSecondary + NFS: https://github.com/facebook/rocksdb/wiki/Secondary-instance + Does RocksDB support multi-process read access? facebook/rocksdb#908 (comment) (and potentially https://aws.amazon.com/blogs/aws/new-a-shared-file-system-for-your-lambda-functions/)
2. Redis on Flash in the same VPC, via ZRANGEBYLEX (e.g. https://github.com/crux-labs/crux-redis)
3. CockroachDB - it supports follower reads via AS OF SYSTEM TIME, and CockroachCloud has VPC peering on GCP
4. DynamoDB+GSI w/ DAX - https://stackoverflow.com/questions/21794945/dynamodb-scan-in-sorted-order/28463257#28463257 (though might be restricted to <10GB https://docs.amazonaws.cn/en_us/amazondynamodb/latest/developerguide/bp-indexes-general.html#bp-indexes-general-expanding-collections)