The Ethernet Alliance will demonstrate a major ethernet interoperability demonstration for supercomputing at the SC12 conference next week.
Supercomputing – or high-performance computing (HPC), as some prefer to call it – has thus far relied mostly on specialist low-latency technologies such as InfiniBand for linking and clustering processors.
The demo seeks to show that today's high-end Ethernet, running at InfiniBand-like speeds of up to 40Gbit/s, can actually be a better choice for HPC, said John D’Ambrosia, who is chairman of the Ethernet Alliance and chief Ethernet evangelist in the CTO office at Dell.
“Ethernet is uniquely able address a variety of demanding environments ranging from supercomputing to today’s converged data centres,” D’Ambrosia asserted. “With its continued evolution and proven flexibility, reliability, and cost-effectiveness, Ethernet is ideally positioned to fuel the next generation of supercomputing applications and advancements.”
Chauncey Schwartz, who chairs the Ethernet Alliance's marketing committee and is in his day job a senior technical marketing manager at networking company Qlogic, added that the demo would also highlight the ability of converged Ethernet to replace three technologies with one, as alongside general networking and HPC clustering, it can also be used as a storage network.
“Big Data, Hadoop processing, MPI [message processing interface, a standard for parallel processing], automated design applications, storage – all of these very intensive, high-performance computing applications need robust, reliable, low-latency connectivity,” he said.
“The demo exemplifies Ethernet’s superior interoperability with the integration of multiple diverse technologies such as 10Gbit Ethernet (GbE) and 40GbE, converged Ethernet with FCoE [Fibre Channel over Ethernet] and iSCSI, and 10GBase-T into a single HPC environment. The demo allows us to illustrate Ethernet’s flexibility, interoperability and capacity for sustained, optimum performance even in the most computing-intensive situations.”
He added that the demo simulates traffic volumes, patterns, and flows commonly found in HPC and data centre environments. It features three layers, including 10GbE-attached servers in the access layer, linked to multiple 10GbE switches in the distribution layer, aggregated by two 40GbE core switches. The equipment used comes from a dozen or so Ethernet Alliance members.
SC12 takes place in Salt Lake City, Utah, from 12-15 November.