Comment by etaioinshrdlu

It looks like it drops the single-threaded limitation of redis to achieve much better performance.

Could this architecture be extended to scale across multiple machines? What would be the benefits and costs of this?