If you need range scan queries (selecting objects and their associations that match certain prefixes), the only DBs that can readily handle this are Bigtable (which is originally designed to store crawldbs) clones like HBase or Hypertable, the latter can do >1M inserts/scans/s sustained on cheap Dell nodes with 16GB ram and 7.2KRPM SATAs. Most KVs based on DHT _cannot_ do efficient scanning. They typically cannot do fast inserts when RAM to DB size ratio is less than 1, either.