Kai

Kai is an open-source distributed key-value database, which is inspired by Amazon's Dynamo. Same as DynamoDB, it is developed for high scalability and validity of Web sites. [goo home](http://home.goo.ne.jp/), a Japanese social networking service with more than 10 million users, is using Kai as the main backend database. Because it is hard to get the search result of the Kai database through the current Web search Engine. It is suggested to search Kai with another keyword, such as Erlang, Sourceforge, or Dynamo.

History

The Kai open-source project started in May 2008 and hosted on [sourceforge.net](https://sourceforge.net/projects/kai/). Till May 2009, 4 versions were released and the latest version is [Kai-0.4.0](https://sourceforge.net/projects/kai/files/kai/0.4.0/). Most of the developers involved in the project were from Japan and the name Kai comes from the village name of one of the developers' hometown.

Data Model

Key/Value

Kai uses a key/value data model. Any value(bit-string) can be stored with a key, which is used to retrieve the value.

Parallel Execution

Bushy

The distributed structure makes Kai's requests to be distributed to N nodes in parallel. Kai uses the actor model, which supported by Erlang, to collaborate each process with each other. One advantage of actor model is that it does not rely on shared memory, which often causes painful bugs.

Query Interface

Custom API

Kai supports basic operations(not all) of memcache API, including get(s), set, delete, quit and version. However, users cannot specify the operations' expiration time, all operation results is persistent.

Storage Architecture

Disk-oriented

Same as Amazon DynamoDB, Kai provides consistent storage in cache. It uses disk-oriented storage architecture.

System Architecture

Shared-Nothing

Kai's system architecture is based on a distributed hash table, a peer-to-peer distributed system architecture. The storage capacity of the cluster can be roughly determined by the number of nodes in the cluster, so the distributed structure enables the system to scale up and down based on the required capacity. To organize cluster nodes, epidemic protocol is used to spread the membership information.

People Also Viewed