Aviz Networks details ONES 3.0 RoCE monitoring and QoS features
Aviz Networks’ ONES 3.0 update expands RoCE and QoS management for AI fabrics, adding watchdog-based fault tolerance, traffic scheduling and WRED queue handling, and a UI focused on real-time plus historical visibility for RoCE, PFC, and congestion events.
Research Overview
The post describes ONES 3.0 as a network management platform feature set aimed at AI/ML data centers and high-performance computing environments using RoCE-based lossless communication.
It frames the update around reducing congestion and loss for AI workloads through enhanced RoCE capabilities, QoS configuration visibility, and monitoring and alerting functions in the ONES interface.
Key Findings
ONES 3.0 adds PFC Watchdog (PFCWD) to detect and address flow control misconfigurations or failures intended to prevent traffic stalls under heavy loads.
The update also includes a Scheduler and WRED for queue and traffic handling, plus QoS features such as DSCP mapping and dot1p mapping tied to traffic prioritization.
Technical Breakdown
The post states that ONES 3.0 builds on ONES 2.0 RoCE support that included lossless communication and proactive congestion management using RoCE over Converged Ethernet.
For RoCE management, it describes UI transparency in a RoCE tab covering DSCP mapping, 802.1p mapping, WRED and scheduler profiles, and PFC plus PFC Watchdog monitoring intended to support lossless RDMA traffic.
QoS configuration and centralized views
The update includes a centralized QoS view that unifies QoS configuration elements such as DSCP-to-TC mappings, WRED/ECN, and scheduler profiles into a single interface.
The post says this UI-based approach reduces reliance on manual CLI access and supports live visibility into queue and profile status for troubleshooting.
Performance and observability in the UI
In the UI, the post describes real-time display of packet rates plus errors and discards, with the ability to access real-time and historical data for network optimization tied to AI workload requirements.
It also outlines time-based monitoring options spanning live views through longer historical windows, and it links those views to troubleshooting and capacity planning use cases.
Operational Impact
The post describes a topology view that provides an interactive map of network devices and connectivity, targeted for monitoring AI/ML and RoCE environments.
It reports features including real-time device status, fault detection for items such as faulty fans, power supplies, or downed links, detailed device metadata, link details for dependency tracking, traffic monitoring for link bottlenecks, and filtering by topology type, device status, and regions.
Leadership Perspective
The post presents the ONES Rule Engine as an enhanced monitoring and alerting experience for proactive anomaly detection across RoCE metrics and AI-fabric counters such as queue counters, PFC events, packet rates, and link failures.
It states that the rule engine supports custom thresholds and conditions for congestion detection and includes alerting integrations with Slack and Zendesk for notifications when anomalies occur.
Overall, the blog frames ONES 3.0 as an end-to-end management and observability update for RoCE-based AI fabrics, combining watchdog-based flow control monitoring, QoS and queue handling features, UI-based configuration and topology visibility, and a rule engine for alerts tied to RoCE and congestion indicators. Blog Signals brief is a fact-based summary of the vendor blog.