Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Flow-Based Telemetry for AI-ML Data Centers

Flow-Based Telemetry (FBT) for VXLANs on QFX5240 switches enables per-flow-level analytics on Integrated Routing and Bridging (IRB) interfaces. It supports EVPN-VXLAN fabrics that use either Centrally-Routed Bridging (CRB) or Edge-Routed Bridging (ERB) overlay architectures. FBT uses Inline Monitoring Services (IMS) to create flows, collect flow statistics, and export flow records to a collector using IP Flow Information Export (IPFIX). FBT can monitor IPv4 and IPv6 traffic in ingress direction, providing detailed visibility into application and workload traffic.

For AI-ML data centers, FBT provides granular visibility into workload traffic flows running across EVPN-VXLAN fabrics, which are commonly used to interconnect GPU clusters, storage systems, and AI applications. This enhances observability beyond what conventional interface counters can offer.

Benefits of FBT for VXLANs in an AI-ML data center:

By exporting per-flow telemetry from IRB interfaces, operators can:

  • Identify high-bandwidth elephant flows consuming fabric resources.

  • Correlate network traffic behavior with AI-ML application performance metrics, to identify how the network impacts AI workload efficiency and responsiveness.

  • Optimize network operations, by identifying overloaded network segments and uneven traffic distribution across the VXLAN fabric.

  • Improve operational visibility for multi-tenant AI environments.

  • Accelerate troubleshooting and root-cause analysis.

  • Support capacity planning based on actual flow-level traffic patterns.

Table 1: Flow-Based Telemetry for VXLANs
   
FBT for VXLANs Overview Learn how to configure flow-based telemetry for VXLANs for the QFX Series switches.