Depending on the characteristics you mentioned (JSON Documents, key access, 100 million records, balanced read / write) I would say CouchDB or Membase are good candidates (here's a quick comparison )
Both HBase and Cassandra may also work, but for HBase you will need to install many components (Hadoop, ZooKeeper, etc.) that you will not use only d, and Cassandra is better if you have more entries, than read it (at least the last time I used it).
The large table, unfortunately, is internal to google :)
source
share