nosql

Installation, testing, benchmarking, and interfacing to NoSQL databases.

There is also a wiki for this repository which will cover testing and performance results of the various NoSQL databases that we explore (minimal content now).

Redis

We (ExMatEx) have been using redis since the early days of the project. At the time, redis did not support distribution. It did support replication for fault tolerance, but not distributing (or sharding) data across multiple nodes to support large databases. With the arrival of redis 3.0, distribution is now supported. This repository contains some instructions and simple test code designed to explore redis 3.0 on LANL's Darwin cluster.

Installation of redis is relatively straightforward. In the past we have installed it as a local module on Darwin. This repo uses redis as if it were installed by a user into their personal directories. It's unlikely we'll be able to install system-wide on other DoE clusters, so testing user installations is essential.

Installation

Redis is available at redis.io. Download, untar, and install the latest version of cluster redis (version 3.0+) according to the instructions on the page. There is a tutorial document that describes how to set up a redis cluster, but this repository contains a script, specific to the Darwin cluster, that will build a redis cluster to support ExMatEx applications.

Requirements

Redis easily installs with no requirements for additional software. However, to set up and run a distributed cluster, Ruby is required. Moreover, the redis gem is also required. We've installed it globally on Darwin, but a savvy Ruby user should be able to figure out how to install it locally using rvm.

Darwin-specific script

The start_cluster_redis.sh script in the scripts directory will build and initialize a distributed version of redis on Darwin. The script steals material from Christoph's original start_redis script in SkelMM and from the redis create-cluster script.

Edit the script and update the redis_tmp and redis_bin variables to reflect where to store temporary files and the install location of redis respectively. The script is set up to use N nodes of the Darwin cluster (which you must have previously allocated using slurm). The script creates N master instances of redis on N nodes and N slaves for these masters on the same set of N nodes. So, you'll have a total of N nodes running 2*N instances of redis. The script supports one command line parameter: start, create, or stop. The following shows the sequence of commands to start the cluster:

salloc -N 3 -p bigmem
cd scripts
./start_cluster_redis.sh start     # start N instances
./start_cluster_redis.sh create    # link them into a cluster
# connect to any port using redis-cli and play
./start_cluster_redis.sh stop      # stops all N instances

An initial port number is specified in the script. This port number is incremented sequentially to address all the masters and slaves. Use any port to connect to the database--queries will be automatically rerouted the node containing the data for the specified key (the cluster redis document describes the hashing algorithm and how a smart client would cache the mapping between keys and nodes).

Finally, the script is not totally automated. Redis software requires that you accept a potential cluster configuration by typing yes to a prompt. There may be a flag to force this but we haven't found it.

Copyright and License

Los Alamos National Security, LLC (LANS) owns the copyright to nosql, which it identifies as LA-CC-2012-065 (ExMatEx: Scale-Bridging Materials Evaluation and Test Suite, Version 1). The license is BSD-sh with a "modifications must be indicated" clause. See LICENSE.md for the full text.