I've abandoned this project indefinitely because it's going to take much more time than I have. The existing work is dubious also, there were some questionable implementation decisions due to noobishness. -Dave

======================

Megalon is intended to be an approximate clone of Google's Megastore database 
for geographically-distributed scalable fault-tolerant consistent data storage.

The paper describing Megastore can be found at:
  http://www.cidrdb.org/cidr2011/Papers/CIDR11_Paper32.pdf

The Megastore paper authors are Jason Baker, Chris Bond, James C. Corbett, 
JJ Furman, Andrey Khorlin, James Larson, Jean-Michel Leon, Yawei Li, Alexander 
Lloyd, and Vadim Yushprakh. All the credit for the interesting protocols goes
to them.

The database is broken up into small pieces called entity groups. ACID
transactions can be run within a single entity group. Each entity group has
a replicated write-ahead log that clients append to using the Paxos
distributed agreement protocol.

This design solves (sort of) the throughput scaling problems of more 
traditional databases that allow transactions over the entire database. The 
Megastore design is not suitable for all applications, though. Round-trip
latency is incurred on writes, and we make the assumption that the chance
of contention is small for any given entity group.   

HBase is used for the underlying data storage due to its similarity to
BigTable. The key feature is MVCC, so clients can read old values while new
values are being written.

Right now megalon is an academic curiosity. If I can avoid screwing it up it 
may someday be fit for production use.

-Dave Revell