Springfield is "big rolla".

It takes the rolla storage engine design and scales it up for use in larger 64-bit systems with millions of keys and data files on the order of GBs+. Like rolla, it is designed assuming it runs on ~100us random access times of solid state disks.

Unlike rolla, it remembers buckets counts in db files, so you don't need to walk on eggshells making sure you use the same NUM_BUCKETS every time with a given db file.

Springfield is fully thread-safe. In fact, You can also compact and "upgrade" the db to higher bucket count while remaining online to get your performance back when the keyspace grows. And, on SSDs, when the page cache is not large enough to cover you, parallel gets on separate threads speed things up nearly linearly.

Springfield also uses CRC sums to validate data integrity of keys/values on disk.

Springfield is alpha. There is at least one known bug/race in online compaction, but everything else seems to work fine.

Still, don't use it in production, yet!

It is pretty fast on SSDs. It outperforms Tokyo Cabinet's TCHDB on most workloads, and often by a wide margin as key counts and db sizes go up.