Kai

Key Features

* Kai is a distributed data store consisting of a cluster of several processes called nodes. It is based on peer-to-peer architecture, and so does not have a single master node, which would be a single point of failure. The storage capacity of the cluster is roughly determined by the number of cluster nodes. The nodes are easily added to or removed from the cluster without interrupting operation. This elasticity of the capacity obviates a prior capacity planning and eliminates up-front capital expense. * Kai is a key-value data store. While key-value stores are recently gathering much attention, they are not well defined. Here, we define them as servers providing a hash table; any value (bit-string) can be stored with a key, which is used to retrieve the value. * In Kai, all data (key-value pairs) are partitioned and replicated among the cluster nodes. The partition and replication rules follow the consistent hashing algorithm. * Kai achieves better latency; it is a couple of milliseconds even in the worst case (of course, it actually depends on data size and configurations). This low latency is achieved by the quorum protocol as well as the parallel processing as described later. The quorum protocol also provides strong consistency model among replicas. * Kai is developed in Erlang, a programming language for parallel processing. This feature is suitable for Kai, since requests of Kai should be distributed to N nodes in parallel for low latency.