![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2023/05/Hemal-Jeff-Laurent-clean.jpg)
![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2023/05/Hemal-Jeff-Laurent-clean.jpg)
Networks are critical as they interlink systems such as servers and storage hardware within high-performance computing systems.
But as the demands of organizations from their HPC systems continue to expand, composable technology allows them to retool the entire architecture as required.
“In the old days when you wanted to build a cluster and you wanted, say, GPS in the cluster, you had to buy a specific server that had GPS in it,” said Jeff Kirk (pictured, middle), engineer at Dell Technologies Inc. “Composable computing brings to bear a very high-speed network that allows you to decide whether that server is a GPU server or a memory server. It’s this concept of an external, very high-speed, low-latency fabric that lets you decide what your server’s architecture is.”
Kirk; Hemal Shah (left), distinguished engineer and system, software and standards architect at Broadcom Inc.; and Laurent Hendrichs (right), senior product line manager of high-speed Ethernet adapters and SmartNIC at Broadcom, spoke with theCUBE industry analyst John Furrier at the ISC High Performance event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed the modern enterprise’s unique infrastructure requirements as it pertains to HPC. (* Disclosure below.)
Networking and hardware go hand-in-hand in most scenarios. Thus, hardware improvements are often the antecedents to cutting-edge networking advancements. One of those is RDMA over Converged Ethernet, a networking protocol leveraging Remote Direct Memory Access over an Ethernet connection.
RoCE has improved in leaps and bounds, bringing it closer in performance and latency to the parallel InfiniBand standard while retaining Ethernet’s benefits, according to Hendrichs.
“RoCE has substantially closed the performance gap with InfiniBand,” he explained. “In the past, InfiniBand might have been the go-to technology for a high-performance network, but right now you have the ability to deploy a network with similar performance and latency using Ethernet, thus taking all the benefits that come with Ethernet in terms of standards, as well a software and hardware ecosystem.”
Scalability is a primary issue the enterprise is dealing with, considering the increasing quantity and intricacy of its data. As graphics processing units become more powerful to crunch this data, a higher load is therefore placed on the network to transport it.
“GPUs are taking it to the next level and because of that, they’re able to process more data,” Shah explained. “That puts more demand on the networking bandwidth. Also, the number of nodes in the fabric, they’re being so high, you need capable Ethernet switches to build this kind of large fabric. You also need scalable transport protocols so that you can support a large number of communication endpoints on a server. Finally, you have to tie all of this together.”
The need for increasing computing horsepower from HPC systems is chiefly due to changing workloads, and artificial intelligence and machine learning are the main contributors to that change, according to Kirk.
“I think the real issue is over the last 20 years, workloads have changed and clusters are now running like clouds with multiple tenants,” he said. “But the workload that really seems to cause the most grief is AI/ML. These require that the network operates at the highest levels of efficiency, maybe 95% or better. What that means is that folks like Broadcom have to devise a good end-to-end solution to deliver that high performance.”
Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the ISC High Performance event:
(* Disclosure: Dell Technologies Inc. sponsored this segment of theCUBE. Neither Dell Technologies nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)
THANK YOU