History

In the Beginning Was Hrothgar...

The Amber cluster was the second Beowulf cluster built by the mathematics department. In 1999, Professor Lisette de Pillis obtained a grant that allowed her and some students to build a Beowulf cluster with sixteen 300 MHz Pentium II nodes, communicating across a 100 Mb/s Ethernet switch.

Beowulf Cluster in original location. The Hrothgar cluster.

The Hrothgar cluster (named for the King of the Danes in the original Beowulf story) had a dual-processor “head” node called hrothgar and a dual-processor development system called wiglaf (Beowulf's loyal companion). It was capable of 1.32 GFLOP per second on the High-Performance LINPACK benchmark. A FLOP is a FLoating-point OPeration; 1 GFLOP/s is one billion floating-point operations performed per second.

Beowulf Cluster in new location. In the CS machine room.

The cluster was moved to a new location in the CS machine room in 2002, to make more room for the mathematics department's server systems.

The Hrothgar cluster was maintained by students and ran until mid-2005, when it was decided that the system was too antiquated to be worth maintaining.

...Then There Was Amber

By 2003, Hrothgar's 300 MHz Pentium II machines were showing their age. The cluster still worked fine for teaching students how to write parallel code, but an average desktop machine provided about the same amount of raw computing power, and a high-end desktop system far outpaced it.

At the same time, several computer-science department faculty members had a need for dedicated compute systems. These faculty members pooled their funds and purchased the sixteen PowerEdge 400SC systems in the new cluster. The CS department also paid for a gigabit switch, which allowed the new cluster nodes to talk to one another. The cluster entered limited service in 2004 (in the department's systems administrator's office!), was moved to the department's machine room in July, 2005, and to its final location in the CS machine room in November, 2005. The cluster was retired in 2010, with the machines donated to charity.

The Amber Cluster, Rear View The Amber Cluster.

The new cluster was called Amber after the Roger Zelazny fantasy series, The Chronicles of Amber. with nodes named after members of the royal family of Amber. The nominal “head node” of the cluster was dworkin, named after the dwarfed Prince of Chaos who created Amber from the elemental Chaos and founded its royal family. Unlike the Hrothgar cluster, each node is an independent computer that can act as a head node if need be. This arrangement allows us to maintain the systems more easily (because they're all identical). It also allows us to “split” the cluster into smaller subclusters should that be desirable—we could have two eight-node, four four-node, or eight two-node clusters, or some combination of these.