Connectivity Innovation in the Age of Open Compute – IT Peer Network
As a founding member of the Open Compute Project, Intel has played a key role in driving specifications and innovations together with the OCP community. With continued advancements in compute performance, heterogenous computing, and new and emerging workloads, including AI, innovation through network connectivity is essential.
In 2019, Intel worked closely with the OCP NIC subgroup within the OCP Server Workgroup to finalize and publish the OCP NIC 3.0 specification version 1.0, which defines an alternative to the PCI Express form factor and enables improvements in serviceability, power delivery, and management. We have expanded the Intel® Ethernet Network Adapter portfolio to include new products in the Intel Ethernet 300, 700, and 800 Series based on this specification at speeds ranging from 1GbE to 100GbE. It is also wonderful to see our OEM partners standardizing on this open specification.
Improving Application Response Time Predictability, Lowering Latency
Our advancements continue with the Intel Ethernet 800 Series Network Adapters, which are in use by select customers now with broad availability targeted for Q3’20. Application response time predictability, latency, and throughput are improved using an advanced traffic steering technology called Application Device Queues (ADQ), a key feature of Intel Ethernet 800 Series.
ADQ has shown significant acceleration of applications including open source Redis, Aerospike and networked storage with NVMe over TCP. Today, I am pleased to share that ADQ is delivering up to 60% improvement in predictability and up to 60% lower latency for a major caching application—memcached—improving service providers’ ability to meet service level agreements (SLAs).*
Co-packaged Optics to Speed I/O performance
Ethernet switches are another critical area for innovation in high-speed connectivity. To address the need for high performance and long-distance data center switch connectivity, Intel is shipping 100Gb/s Silicon Photonics transceivers in high volume today, with 200Gb/s and 400Gb/s transceivers ramping in 2020.
As we look to the future, switch ASIC bandwidth is continuing to grow to keep up with data center traffic. The I/O performance and power efficiency, however, are not keeping pace, and continued bandwidth scalability is facing challenges from SERDES and copper I/O limitations. To answer these challenges and to enable continued scaling, we can move the optical network interfaces inside the switch package. This reduces the power consumption of the switch and associated interconnects. It also removes the need for retimers, which add cost and power.
To address these needs, the Co-Packaged Optics Collaboration (CPO) was announced at the OCP conference last year with a goal to foster a multi-vendor ecosystem through the definition of common standards for the switch system building blocks and for the interfaces between them. As illustrated by our industry-first co-packaged optics Ethernet switch, Intel is at the forefront of developing the technology to enable the integration of optics and networking silicon in the same multi-chip package. We remain firmly committed to open standards and development of a vibrant ecosystem as a contributor to the CPO consortium.
Fueling Innovation in Data Center Switching with Tofino and SONiC
In partnership with Microsoft, our Barefoot Division is updating on continued contributions to SONiC. Barefoot has been active in the SAI and SONiC community with support for multiple Tofino-based white box switches. Barefoot is extending SONiC support to its next generation, 12.8Tbps Tofino 2 Ethernet switch ASIC. The P4 programmability of the Tofino family of ASICs enables differentiating features like real-time data plane telemetry and flexible table scale.
Barefoot has also partnered with Keysight Technologies, who announced in March support of a new, cost effective, dev-ops friendly multi-terabit network tester platform based on Tofino, which will help accelerate verification of high bandwidth network switches.
Supporting an Open Hardware Movement
Since OCP’s founding, Intel has driven more than thirty contributions and enablements with over one hundred products and partners spanning compute, network, and storage. We at Intel recognize that open standards are key to our goal to deliver value through an open ecosystem across the full platform lifecycle and are proud to support OCP and the open standards it is fostering to better enable hyperscale data centers.
I look forward to continuing our broad partnerships to drive the necessary innovations in connectivity to meet the burgeoning demands on the network. To learn more about our OCP contributions, don’t miss Rebecca Weekly’s blog on Intel’s broader OCP engagement and visit https://www.opencompute.org/contributions.
Written by Steve Schultz, Vice President of the Connectivity Group & Director of Business Execution Ethernet Networking Division
* Results have been estimated using internal Intel analysis and provided to you for informational purposes. Any differences in your system hardware, software or configuration may affect your actual performance. For more complete information about performance and benchmark results, visit www.intel.com/benchmarks.