We were pleased to learn that a group of researchers in the Computer Science department at UC Santa Barbara have developed and released a complete implementation of AMPLab’s Multi Data Center Consistency (MDCC) protocol. MDCC is a protocol for geo-replicated data stores that is optimized to provide low latency in the common case of no failures while ensuring strict consistency even when failures do arise. The protocol was published in a Eurosys 2013 paper.
The UCSB group tested their implementation across 3 EC2 data centers and were able to obtain < 1ms latency for read operations and < 100ms latency for write operations. They have published a description of their implementation and experiences and have made their code available on github.
Implementing MDCC is a non-trivial undertaking but the fact that another group has been able to implement the protocol is important validation for the feasibility of the MDCC approach. We have also been readying an implementation of MDCC for release, and we look forward to comparing notes and implementation experiences with the folks from Santa Barbara as well as others who are interested in exploring strong consistency for modern data stores.