Hardware Reference
In-Depth Information
Build a Raspberry Pi Supercomputer
HACK 65
There seem to be two things people immediately want to do with their
Pis: make a media server (see
Hack #54
) or make a cluster of Raspberry
Pis. This hack is for those of you in the latter group.
Be þéos sceaft, sum clyster, nemnende for se eponymous ceorl fram… Just kidding.
We're talking a Beowulf
cluster
, not the Old English stuff. And not Old English like the
wood cleaner, like the epic poem. Either way, it doesn't matter. By the time we trans-
lated this whole hack into mediocre Old English, our grandchildren would be writing
about installing Fedora 140 on the Lingonberry Muffin (all the rage in 2073—keep an
eye out).
A Beowulf cluster is what happens when you connect some smaller computers (most
likely some cheap bits somebody stuck in a storage closet a while back, or as Wikipedia
more politely describes it, “commodity-grade”) together to build a supercomputer.
The term came from such a computer built at NASA in the mid '90s, so named because
the eponymous character of
Beowulf
had “thirty men's heft of grasp in the gripe of his
hand.”
The following instructions are adapted from those written by Simon Cox in the Com-
putational Engineering and Design Research Group at the University of Southampton
for a
Raspberry Pi supercomputer with LEGO racking
. It uses MPI (Message Passing
Interface) to communicate between nodes.
These instructions assume you are using Raspbian. When you first boot the image,
expand the image to fill the card. Don't forget to change the default password and
refresh the list of packages in your cache (
sudo apt-get update
). Then you're ready to
start supercomputing!
Background Reading
Parallel Processing on the Pi
is an excellent post to read, if only for reassurance
that you are going to make it by the end, but don't start following those instructions.
You're going to build everything yourself in this hack.
You'll also want to look at the
MPICH Installer's Guide
to better understand some
of the steps in this hack.
Build MPI to Run Code on Multiple Nodes
To begin, get Fortran:
$ sudo apt-get install gfortran
After all, what is scientific programming without Fortran being a possibility?
