Peloton (supercomputer)

The Atlas cluster

ThePelotonsupercomputer purchase is a program at theLawrence Livermore National Laboratoryintended to provide tera-FLOPcomputing capability using commodity Scalable Units (SUs). The Peloton RFP defines the system configurations.

Appro was awarded the contract forPelotonwhich includes the following machines:

Machine	Nodes	TPP (TFLops)
atlas	1152	44.24
hopi	80	2.92
minos	864	33.18
rhea	576	22.12
yana	80	3.07
zeus	288	11.06

All of the machines run theCHAOSvariant ofRed Hat Enterprise Linuxand theMoabresource management system. Under the project management of John Lee, the team at Synnex, Voltaire,Supermicroand other suppliers, the scientists were able to dramatically reduce the amount of time it took to go from starting the cluster build to actually having hardware at Livermore in production. In particular, it went from having four SUs on the floor on a Thursday, to bringing in two more SUs for the final cluster and by Saturday, having all of them wired up, burned in, and running Linpack.