Cisco is adamant about reinforcing Ethernet as the foundational pillar for AI-focused networks, both now and in the foreseeable future.
Their significant involvement in Ethernet’s evolution through IEEE and other industry alliances has been noteworthy. Presently, they are at the forefront of the Ultra Ethernet Consortium (UEC), a team striving to bolster Ethernet across various layers to suit AI-driven architectures.
“Considering the vast data businesses hold today, there’s a growing interest in AI,” remarked Thomas Scheibe, senior director at Cisco’s Cloud Networking division. “They’re keen on understanding the networking adjustments necessary to manage the expansive data clusters and processing demands, with Ethernet emerging as the go-to solution,” Scheibe explained.
To navigate this, Cisco has crafted a roadmap to guide enterprises on harnessing their current Ethernet-based data centers for immediate AI tasks.
Enhancements in Nexus 9000 Series Central to Cisco’s AI schematic is the Nexus 9000 series, which can handle immense bandwidth. These switches are equipped to cater to AI/ML needs, ensuring minimal latency and efficient congestion management.
The Nexus switch’s key features for AI-centric networking include its NX-OS operating system, which supports the Remote Direct Memory Access Over Converged Ethernet, version 2 (ROCEv2) and Explicit Congestion Notification (ECN). ROCEv2 enhances data transmission speeds, while ECN ensures uninterrupted network flows, a must-have for AI-centric tasks and even real-time voice or video communication.
Additionally, the Priority Flow Control feature aids in managing congestion in advanced networks.
Scheibe added, “We’ve equipped Nexus with technologies that prioritize specific tasks, ensuring that AI tasks with zero tolerance for packet loss always get top priority.” Cisco also offers scripts for seamless network configuration and optimization.
Moreover, the Nexus 9000 series offers advanced telemetry features for analyzing network performance, a crucial aspect for RoCEv2 optimization.
Beyond Nexus: The Silicon One Processor Another pillar of Cisco’s AI network strategy is its advanced Silicon One processors, targeting extensive AI/ML infrastructures.
Cisco’s latest additions to its Silicon One series emphasize a unified silicon architecture for both routing and switching, supported by a standardized operating system and SDK.
These devices are optimized for AI/ML needs, boasting advanced Ethernet features, ensuring fluid traffic flow, minimized congestion, and lowered latency.
Cisco’s concept of a “Scheduled Fabric” amalgamates these Ethernet advancements, creating a synchronized communication network for elevated bandwidth efficiency.
Addressing Data Center Sustainability Beyond AI, sustainability in data centers is gaining traction. Scheibe mentioned challenges like efficient network scaling and sustainable operations. “Companies aim to comprehend their current energy consumption to make sustainable future decisions,” Scheibe noted.
With services like Cisco’s Network Energy Utilization and recent integrations in the Nexus Dashboard, businesses can access real-time energy consumption data, promoting more sustainable operations.
Furthermore, the Nexus Dashboard offers an AI Data Center Blueprint for Networking, assisting businesses in optimizing their networks for AI-intensive workloads, including strategies for advanced network migrations and the establishment of expansive machine-learning frameworks.