JRS: Job-Running System

JRS, or "Job-Running System", is a very basic batch system -- that is, a suite of network software that can spawn "jobs" on a cluster of "compute nodes". In this context, a "job" is just a command line -- for example, a simulator or a script to perform a mathematical computation -- and a "compute node" is just an ordinary Linux machine sitting on a network. JRS does not presume to know about any other infrastructure -- for example, it does not know about any shared filesystems or any network authentication systems. Its design is intentionally simple: its only purpose is to dispatch command lines to remote hosts.

JRS is a (distant) little cousin to mature systems such as Condor and PBS. It differs from these systems in that it is extremely small and very simple to set up.

JRS has a homepage. Its source repository is hosted on GitHub. Packages for Debian-based distributions can be found online.

Installation

Compiling

1. Install OpenSSL and APR (Apache Portable Runtime). On Ubuntu/Debian, these packages are libssl-dev and libapr1-dev.

2. Install the scons build tool if you do not already have it.

3. Compile with scons. This will produce a static binary jrs suitable for distribution to a compute cluster without the need to install the dependencies on each node.

Installing

JRS can be installed either as a Debian package or "from scratch". To build a Debian package, do the following:

$ ./build-deb.sh
$ dpkg -i jrs_1.0-1_amd64.deb   # for example

Alternatively, simply do make install to install JRS.

The JRS daemons should be started on boot for an ordinary compute cluster configuration. JRS installs an init-script at /etc/init.d/jrs-daemon. This script can be used to manually start/stop/restart the daemon(s) as well.

Configuration

JRS requires one "metadata server", which coordinates resource sharing among all clients, and one or more "compute nodes", which actually run jobs. To configure JRS, first determine which nodes should serve each of these roles.

1. Configure the master by editing node.master in the JRS installation directory. This file should be updated on all nodes if the master is changed.

2. Configure the compute node list by editing node.list in the JRS installation directory. This file should also be propagated to all nodes.

3. Set up a shared secret among all machines. This shared secret is used to establish encrypted communcation between clients, the metadata server and the compute nodes. Place random data of arbitrary length in the secret file and propagate it to all nodes. For example:

   dd if=/dev/random of=secret bs=256 count=1
   

Using

Once JRS installed, its use is very simple. Just use jrs-submit with a Condor submit file:

$ jrs-submit test.condor

The submit process will run, and show status updates, until all jobs are done. If the process is killed and restarted, it will restart all jobs, so make sure to run it on a node that will remain stable.

Authorship and License

JRS is fully original code Copyright (c) 2013 Chris Fallin cfallin@c1f.net. It is released under the GNU General Public License, version 2 or later.