News

SURF’s innovation department recently collaborated with Cornelis Networks to advance networking capabilities for high-performance computing (HPC) clusters. The collaboration marks a new step forward for SURF in optimizing HPC and artificial intelligence (AI) applications while enhancing network performance and efficiency.

Collaboration background

Cornelis Networks is a provider of cutting-edge network hardware solutions. With a focus on innovation and performance, Cornelis develops Omni-Path network hardware, positioning itself as a strong competitor to established interconnect standards. Through their commitment to delivering high-speed, low-latency networking solutions, Cornelis aims to revolutionize the landscape of networking for research computing.

Cornelis Omni-Path Express delivers low latency and high message rate, the foundations for application performance and scalability. — Cornelis Networks.

Goal partnership

The collaboration between SURF’s innovation department and Cornelis Networks began approximately two years ago at ISC Hamburg, where discussions about testing Cornelis’ network hardware within SURF Experimental Technologies Platform first took place. Following subsequent conversations and evaluations, SURF and Cornelis Networks agreed to initiate a Proof of Concept (PoC) during SC23 last fall.

One of the key advantages of Cornelis’s solution is its easy integration with existing systems. The installation process was facilitated by kernel drivers and software included in the Debian OS. With the assistance of SURF’s configuration management engine, the complete software stack was deployed efficiently, enabling quick and hassle-free implementation. — Jaap Dijkshoorn, SURF Experimental Technologies Platform.

As part of the PoC, SURF received a comprehensive set of Cornelis hardware, including a 48-port switch (100SWE48UF2), HBA cards (CN-100HFA016LSN), and Omni-Path Active Optical QSFP cables. These components were integrated into SURF’s Experimental Technologies Platform, including various computer servers and in Liqid decomposable infrastructure fabric chassis.

The key aspect of this collaboration is introducing Cornelis as an HPC interconnect option. This will allow researchers and engineers to use Cornelis versus other existing solutions. The Dutch scientific computing community can leverage the systems in Experimental Technologies Platform for academic purposes and contribute to future PoCs.

Initial benchmarks, including OSU benchmarks, yielded promising results, validating the anticipated performance improvements. Moving forward, the focus will shift to real-world applications, leveraging SURF’s infrastructure to assess the full potential of Cornelis’ networking solution for HPC clusters. Additionally, plans are underway to explore job execution within the Liqid decomposable infrastructure, further optimizing research workflows.

Interested?

SURF Experimental Technologies Platform has a dedicated team of expert system administrators and advisors. Are you interested in co-creating and experimenting with emerging technologies and methodologies? Then share your innovative challenges with us!