Hive

A collection of distributed algorithms. See subdirectories..

chord

Distributed hash table.

memcached

An implementation of the memcached protocol built mainly to excercise the chord algorithm. It is slower than THE memcached by about a factor of 10. And it doesn't fully implement the whole protocol. Currently, it implements just the binary protocol enough to make memslap think it's talking to a real memcached server.

more to come...

HOWTO

Chord And Memcached are both Erlang 'application's. Chord can be used stand alone from within Erlang. To provide an external interface to Chord, I decided to implement a Memcached server. Memcached depends on Chord. And only one instance of Memcached should be started in a machine because the sole purpose of Memcached is to bind to port 2222 and on any machine you can only have one application bind to a port. You can start multiple instances of Chord even within the same physical machine if you run multiple Erlang nodes.

It's not completely polished so there are some manual steps to get it up and going:

1. Start 4-5 terminal sessions.
2. On each terminal run erl -sname nX@localhost where X is a number that is distinct on each terminal (Important: see Node Names).
3. After starting erlang on all the terminals, at the erl> prompt, type the following set of commands:

    erl> net_adm:world().
    erl> application:ensure_all_started(chord).
    erl> chord_server:join().

4. On one of the erlang nodes, type:

    erl> application:ensure_all_started(memcached).

You should now have memcached up and running. The memcached server uses chord to store and retrieve values from the chord ring. Chord, in turn, uses ETS for storage. The code is designed to be able to swap it out for DETS or filesystem storage or what-have-you.

See chord to play with the chord ring.

See memcached to play with the minimal memcached implementation.

Important

there is a .hosts.erlang file that says to look for other nodes only on localhost. If you want to try it out on different physical nodes, be sure to edit the .hosts.erlang file or otherwise use the appropriate net_adm:ping/1 commands instead of net_adm:world/0. In the future, I'll probably integrate some method to auto-discover nodes running chord.