Gigabit Ethernet cluster gets accolade

High placing on supercomputer list earns praise

Gigabit Ethernet as a means to connect devices in datacentres and storage facilities has earned another proof point — a computing cluster in Germany that is linked by Gigabit Ethernet has ranked number 58 overall for performance among the top 500 supercomputing sites in the world.

The cluster at Max Planck Institute in Hanover, Germany, supports a 32.8 Teraflop cluster, which is a far cry from the petaflop level reached by IBM to claim the top spot on the Top500.com list of supercomputing installations.

But it is still an impressive amount of computing power generated by off-the-shelf equipment, and ranks the site as the number one Gigabit Ethernet-connected facility on the list from among 285. That makes Gigabit Ethernet the second most popular connecting technology among the top 500, behind Infiniband, which was used in 120.

The winning IBM supercomputer uses both Gigabit Ethernet and Infiniband.

The Max Planck cluster, called ATLAS, consists of Intel EM64T 32xx 2.4GHz processors supplied by Pyramid and connected via Gigabit Ethernet switches made by Woven Systems.

The switches feature Woven technology called Dynamic Congestion Performance. This detects when traffic bound for the same computational device will collide with other traffic with the same destination and switches some of the traffic to an alternate path.

This effectively prevents switches that connect the computational devices from being overwhelmed. It also reduces the amount of memory needed to buffer the switches because they have fewer collisions to deal with. Less memory lowers the cost of the switches, according to Woven Systems.

ATLAS crunches numbers in an attempt to directly measure gravitational waves, the warping of the space-time fabric that was postulated by Albert Einstein in 1916 but never directly observed.

Join the newsletter!

Error: Please check your email address.

Tags technologygigabit ethernetatlasmax planck institute

Show Comments
[]