kavy

Networked in-memory key-value store, written in pure C.

How does Kavy work?

Kavy runs on an event-loop architecture, which works as follows

• The server waits until any clients are readable or writable, checking each one
  • If there is input, the server reads and appends to client->input
  • If there's enough data to run commands it does so, appending the output to client->output
  • If there is data in client->output, the server sends as much as possible, shifting the sent data from the beginning of the output buffer

What makes Kavy different from Redis?

• Binary protocol - speed is prioritised here, rather than compatibility
• Static hash table with a fixed amount of buckets (1024^2 by default, which takes up only 8MB on a 64-bit system)
• Literally only two features - set and get

What's the performance like?

In my synthetic benchmarks, Kavy was up to 10 times faster than Redis - likely due to it's pure simplicity and lack of logging and client-specific error handling.

Protocol

All values are big-endian.

set

Request

uint8 1 uint32 key_length uint32 value_length char[] key char[] value

Response

uint8 0 or uint8 1 for fail/success respectively

get

Request

uint8 2 uint32 key_length char key[]

Response

uint8 0

or

uint8 2 uint32 value_length char[] value