Thank you for your message.
Hmm, it is difficult topic. Since the design of kumofs is completely different from other distributed storages like Voldemort or Riak, my thought is that kumofs' characteristic feature is distributed storage with strong consistency. Thus kumofs will support different use cases from these eventually-consistent based distributed storages, for now and for the future.
The most important plan is improvement of data model. Present key-value model is simple but too tiny for many applications. Applications can build rich data structures on the top of kumofs using CAS operation, but the thruput will be slow. Rich data model will improve these performance drastically. The model will be queue, map or table with query language (possibly SQL :-) ) as kumofs supports strong consistency.
The plan will be done with MessagePack-RPC. It is another my open-source project to create scalable messaging system (in other words, better Thrift or better Protocol Buffers). I'm working on it now. And now the development of kumofs is "bug-report driven" or "feature-request driven". The time of principal feature improvement will be after the MessagePack project.
Thanks,
Sadayuki
--
Sadayuki Furuhashi
frs...@gmail.com