You are hereHistory

History


LittleFe began as an idea by Paul Gray (University of Northern Iowa), Dave Joiner (Kean University), Tom Murphy (Contra Costa College), and Charlie Peck (Earlham College) in 2005. While they had been teaching computational science education, they realized that their curricula depended on local computing resources that were not always present.

Over the course of its development, many engineering issues had to be addressed. LittleFe started as a cluster of eight small PC computers in separate cases. Unfortunately, the separate cases, power supplies, and hard drives made the system quite heavy, and was expensive to bring on an airplane. The first step, therefore, was to reduce weight.

The cases were the obvious place to start. The cases were largely cosmetic and easily weighed more than the motherboard itself. Initially we created a simple wooden frame for the nodes, and mounted each motherboard on its own board to slide into the frame. While wood is easy to work with, it is mo re fragile than metal so the final solution was to have a local metalworking firm fabricate an equivalent frame from aluminum.

After much experimentation (involving smoke and flame) with desktop-style ATX power supplies, we settled upon a large shared power supply taking in 110VAC and producing 12VDC. We then attached small micro-ATX power supplies to each node to produce all the other voltages necessary for the motherboard. With the shared power supply mounted on the side of the frame, we ran power cables from a 12VDC lug to the micro-ATX power supplies on each node. Eventually, we soldered barrel connectors onto the nodes so that we could disconnect the power from a node easily.

Finally, we had to wrestle with what software we would provide on the cluster. We started with a basic Debian installation, which we maintained separately for each node. This was not a very maintainable solution, and detracted from the ease-of-use that LittleFe was supposed to deliver. We then started experimenting with running running Paul Gray's Bootable Cluster CD (BCCD) on the cluster. This was much more maintainable, but its live CD nature meant it was difficult to have configuration and data persist between reboots of the cluster.

This was the start of the "Liberation" project for BCCD. Ideally we wanted a single head node with a hard drive, and have the rest of the nodes use the network to run off that drive. The process we settled on involved booting the node with the hard drive into live CD mode, and then running a shell script to copy the contents of the live CD to the hard drive, and then configure that system to boot the rest of the cluster nodes over the network.

In November 2010 the LittleFe project was the recipient of a grant from Intel to build 25 clusters to be given to faculty across the United States who are involved in computational science education. These faculty will use their LittleFe clusters to improve or develop curricula for their students, and ultimately for such globally-available resources as HPC University (HPCU) and the Computational Science Education Reference Desk (CSERD).