Eliminate Monitoring Blind Spots

SNMP polling at 1 to 5 minute intervals captures time-averaged interface utilisation statistics that hide the congestion events that cause application tail latency and packet drops. A micro-burst that fills a switch queue to capacity and triggers packet loss in 200 microseconds leaves no trace in a 5-minute average utilisation graph: the average utilisation may be 30% during an interval that contained a congestion event that caused 1,000 packets to be dropped. When a network operations team investigates an application performance complaint that correlates with the 200-microsecond burst, SNMP statistics provide no evidence that any network event occurred, directing investigation toward application servers or end-user equipment rather than the switch queue where the actual packet loss occurred. The visibility gap exists not because the network lacks telemetry hardware, but because traditional monitoring approaches collect telemetry at intervals many orders of magnitude longer than the timescales at which congestion events occur.

Arista LANZ (Latency Analyzer) captures queue depth and latency at nanosecond precision from every switch buffer in EOS hardware, recording the peak occupancy and duration of each congestion event and streaming that data to CloudVision in real time, without requiring external probe hardware, external tap infrastructure, or separate monitoring appliances. Because LANZ is part of EOS and runs on every Arista switch, the telemetry is available uniformly across all fabric tiers using the same API and data schema. DANZ (Data Analyzer) extends this visibility to include packet-level TAP aggregation for deep inspection and continuous capture, providing the capture infrastructure for security forensics and compliance use cases where historical packet data is required. The 7010X achieves 296 Gbps and 220 Mpps at under 0.3W per Gbps, and the 7260X3 provides 12.8 Tbps in 2RU with a 64 MB dynamic shared buffer and 9.5 Bpps, both instruments for LANZ monitoring. EOS live patching and stateful fault containment keep the monitoring software running without maintenance-window disruptions.