PelicanHPC: A GNU/Linux distribution to create a HPC cluster for MPI based parallel computing

PelicanHPC GNU Linux

UAB

News

PelicanHPC v2.0 is available. Features:

based on Debian testing (squeeze) instead of stable (lenny). This means that most packages have newer versions. In particular, the kernel is at 2.6.30 and Open MPI is at 1.3.3.
has new MPI bindings for GNU Octave, developed mostly by R. Corradini, building off MPITB. The new MPI bindings allow use of Octave 3.2.x instead of 3.0.x, which gives some important performance gains. The new bindings are less complete than MPITB, but they provide all MPI calls used in the examples for GNU Octave. The Monte Carlo and kernel examples have been adapted to use these new bindings, the other examples of MPI usage within Octave still need to be updated.
Open MPI is now the only MPI implementation installed.
the Ganglia monitoring system is installed and pre-configured for up to 4 hosts. It is easy to add entries for larger clusters. Visit http://localhost/ganglia after having set up the cluster (pelican_setup). The ksysguard monitor is still available, too.

The Linux Pro Magazine article on PelicanHPC from the June issue is now available online.
14 September 2009. There is a fast new download mirror, thanks to the Medical Informatics department of Heilbronn University and Heidelberg University. Thanks to Simon Streicher for setting this up!
30 June 2009. I finally managed to make a decent screencast that shows how to use PelicanHPC. It's 34MB, though. Check it here if you have the curiosity and bandwidth.
29 Sept 2008. If you were looking for ParallelKnoppix and you arrived here, you're in the right place. PelicanHPC is the continuation of the ParallelKnoppix project. The PK web page is no longer available, and re-directs here. ISO images for the last version of PK are achived here on the download page, but you're encouraged to use PelicanHPC instead.

Features

PelicanHPC is a live CD image that let's you set up a high performance computing cluster in a few minutes. A Pelican cluster allows you to do parallel computing using MPI. You can run Pelican on a single multiple core machine to use all cores to solve a problem, or you can network multiple computers together to make a cluster. The frontend node (either a real computer or a virtual machine) boots from the CD image. The compute nodes boot by PXE, using the frontend node as the server. All of the nodes of the cluster get their filesystems from the same CD image, so it is guaranteed that all nodes run the the same software. Packages can be added to the frontend node on the fly, thanks to aufs. If new software is installed in /home/user (easy for self-compiled, a little tricky for .debs), it is available to all the nodes. The CD image is created by running a single script, which takes advantage of the Debian Live infrastructure. It is very easy to create a custom version with new packages installed in standard locations by adding package names to the script and then running it.

Pelican is created using Debian Live. To make your own version you only need live_helper (also deboostrap, rsync, any others?) and the make_pelican script, which is provided below.
The LAM-MPI and OpenMPI implementations of MPI are installed. Both 32 and 64 bit versions are available. Debian testing (Lenny) is the base for both.
Contains extensive example programs using GNU Octave and MPITB. Also has the Linpack HPL benchmark.
You can use any Class C network you like. By default, the cluster is on 10.11.12.*
xfce4 window manager, konqueror for browsing and file management, ksysguard for monitoring the cluster, kate and nano for editing. As noted, it is very easy to add packages. Pelican is a bare-bones framework for setting up a HPC cluster.
Pelican releases and all testing is currently done using Debian Lenny as the base. Squeeze or sid may or may not work.

Here are some screenshots (click to enlarge):

Download

I'd appreciate it if you would try to get a new release using bittorrent, if possible. Check for torrents here. When the torrent gets stale, use the normal download page or a mirror (see below). The make_pelican script and pre-built ISO images are always available on the normal download page. To use the script to make your own version, you need to install the live_helper package. Get the most recent version.

Mirrors (Thanks!):

Germany:
http://download.mi.hs-heilbronn.de/pelicanhpc/ Provided by the Medical Informatics department of Heilbronn University and Heidelberg University.

http://download.semmel.org/PK/ username "parallel" password "knoppix"

Documentation

The main documentation is the tutorial, which is a series of screenshots with comments. For those of you who fly by the seat of your pants, boot a computer using the CD, and follow the instructions that come up on the screen.
Screencast video that shows how to set up and use PelicanHPC. Warning - this is about 34MB.
User Forum: the main place to ask questions
Article in Linux Pro Magazine from June 2009.

Highlights from the forum:

Academic work that uses PelicanHPC or ParallelKnoppix

Please let me know if you use this software to produce published academic work.

M. Cech (2009) ŘEŠENÍ PRO CLUSTEROVÁNÍ SERVERU, Masters thesis, BRNO UNIVERSITY OF TECHNOLOGY

Dietlmeier, J. Begley, S. Whelan, P.F. (2008) Cost-Effective HPC Clustering for Computer Vision Applications ,Centre for Image Process. & Applic., Dublin City Univ., Dublin; This paper appears in: Machine Vision and Image Processing Conference, 2008. IMVIP '08. International
Publication Date: 3-5 Sept. 2008

J. Aldazabal et al. (2008) Diffusion simulation of Cr-Fe bcc systems at atomic level using a random walk algorithm physica status solidi (a)
VL: 205, NO: 6, PG: 1337-1342, YR: 2008

Muhammad Sharif, and Aman Ullah Khan (2007) Benchmarking of PVM and LAM/MPI Using OSCAR, Rocks and Knoppix Clustering Tools, PROCEEDINGS OF WORLD ACADEMY OF SCIENCE, ENGINEERING AND TECHNOLOGY VOLUME 21 MAY 2007 ISSN 1307-6884

T. Grytsenko & A. Peratta (2007) Implementation and performance assessment of a parallel solver for sparse linear systems of equations and rules for optimal solution, in Data Mining VIII: Data, Text and Web Mining and their Business Applications

Syarif Hidayat (2006) PEMROSESAN PARALEL MENGGUNAKAN KOMPUTER HETEROGEN.. Seminar Nasional Aplikasi Teknologi Informasi 2006 (SNATI 2006) ISSN: 1907-5022.

Mustafa Çakır, Gonca Çakır, Levent Sevgi (undated) PARALLEL FDTD-BASED RADAR CROSS SECTION SIMULATIONS

Similar distros

Some other distros for setting up a HPC cluster are mentioned here. They may be better choices for certain tasks.

KestrelHPC is another new system for setting up a HPC cluster, using a frontend machine that runs Debian or Ubuntu GNU/Linux. It allows for multiple users, whereas PelicanHPC is for a single user. From the web page: "This project offered what we were looking for (easy to use and an easy interface to configure the cluster), yet, we wanted an installed linux instead of a "live" distribution. Hence, on account of to the aforementioned, we started developing a script that would make a Debian or an Ubuntu, a Pelican-like linux. This consists of, a script that installs all the packages needed to have a working Head-node and besides this, copies the image that will be sent to the slave nodes by LAN. Finally, modified pelican scripts and new kestrel scripts are copied to administrate the cluster in a very intuitive way. "
ABC Gnu/Linux is a new distro for setting up a HPC cluster that works in both live and installed modes. Details and documentation are still sparse. I have tried it in live mode on a virtual cluster, and it works nicely, in my experience. It has the ganglia monitor, which is nice for seeing the details of the operation of the cluser. The frontend node automatically begins to operate as a dhcp server, using eth0, so be careful not to interfere with existing networks if you boot it up. I recommend giving it a try.
I ran across Cluster by Night which is another live CD approach to setting up a cluster for MPI work. It works within an existing network, which is not the case with PelicanHPC. It seems to be a very nice solution when it is not possible to set up a private network.

How much interest is there in PelicanHPC?

This part is here to help convince people that funding this project is worthwhile. Check at StatCounter (then click on Recent Visitor Map) to see an updated version of this map:
user map from StatCounters (thanks!)