Arista Networks, a recognized leader in cloud and AI networking, has unveiled advanced innovations designed to enhance AI cluster performance and efficiency. The latest enhancements include Cluster Load Balancing (CLB) within Arista EOS®, optimizing AI workloads through stable, low-latency network flows. Additionally, Arista CloudVision® Universal Network Observability™ (CV UNO™) now features AI job-centric observability, facilitating more efficient troubleshooting and rapid issue identification to ensure reliability at scale.
Advancing AI Networking Intelligence
The Arista EOS Smart AI Suite is engineered to provide AI-grade resilience and security, empowering AI clusters with the groundbreaking Cluster Load Balancing feature. This Ethernet-based AI load balancing solution leverages RDMA queue pairs to maximize bandwidth utilization across spine and leaf switches. AI clusters typically consist of a few large bandwidth flows, making conventional load balancing methods ineffective, often leading to uneven traffic distribution and higher tail latency. CLB mitigates these inefficiencies by implementing RDMA-aware flow placement, ensuring optimal performance across all flows while maintaining minimal tail latency. By taking a comprehensive, bidirectional approach to traffic optimization between leaf and spine switches, CLB ensures balanced network utilization and sustained low latency.
“As Oracle continues to grow its AI infrastructure leveraging Arista switches, we see a need for advanced load balancing techniques to help avoid flow contentions and increase throughput in ML networks,” said Jag Brar, vice president and Distinguished Engineer, Oracle Cloud Infrastructure. “Arista’s Cluster Load Balancing feature helps do that.”
Also Read: Juniper Networks Launches Solution for GPUaaS & AIaaS
Comprehensive AI Observability
Arista’s AI-driven 360° Network Observability platform, CV UNO, powered by Arista AVA™, delivers end-to-end AI job visibility by integrating network, system, and AI job data within the Arista Network Data Lake (NetDL™). The EOS NetDL Streamer, a real-time telemetry framework, continuously streams granular network data from Arista switches into NetDL, offering an advanced alternative to traditional SNMP polling. Unlike periodic queries that may overlook critical network events, the EOS NetDL Streamer delivers high-frequency, low-latency, event-driven insights into network performance—a key enabler for large-scale AI training and inference infrastructure. Tailored for AI accelerator clusters, it enhances impact analysis, pinpoints issues with accuracy, and accelerates resolution times to ensure seamless AI job execution.
Key benefits include:
- AI Job Monitoring – Provides a detailed overview of AI job health metrics, such as job completion times, congestion indicators (ECN-marked packets, PFC pause frames, packet drops), and buffer/link utilization for real-time insights.
- Deep-Dive Analytics – Identifies critical job-specific issues by analyzing network devices, server NICs (e.g., PFC out-of-sync events, RDMA errors, PCIe fatal errors), and associated data flows to pinpoint performance bottlenecks with precision.
- Flow Visualization – Utilizes CV topology mapping to deliver real-time, intuitive visibility into AI job flows at microsecond granularity, expediting issue detection and resolution.
- Proactive Resolution – Detects anomalies early, correlating network and compute performance within NetDL to ensure uninterrupted, high-efficiency AI workload execution.
With these latest advancements, Arista Networks continues to set the standard for AI networking, delivering cutting-edge solutions that enhance performance, reliability, and efficiency at scale.
