Aviz Networks outlines how ONES measures network latency with scheduled agent probes

The ONES network latency measurement backend adds an agent-based system for SONiC switches and servers that can trigger scheduled ICMP or TCP latency probes and calculate results with nanosecond precision to support network monitoring and SLA validation.

Research Overview

The blog describes a core component within ONES that collects and processes network latency data using endpoint-based agents. It frames the component as the engine that enables latency measurement across an ONES collection ecosystem by collecting results from configured destinations.

The post also outlines how latency measurements are initiated, which protocols are supported, and how periodic execution supports operational needs such as baseline creation and SLA checking.

Key Findings

The backend runs an agent in both SONiC switches and servers and exposes an API through which ONES collectors can trigger latency measurements. Latency to a destination IP is measured using either ICMP or TCP, with TCP requiring a port parameter in the measurement inputs.

The blog emphasizes that scheduled latency calculations can be performed at defined intervals, such as every five minutes, to support monitoring, troubleshooting, capacity planning, and measurements relevant to real-time applications.

Technical Breakdown

The post states that latency measurements depend on protocol choice, destination IP, and, for TCP, the port. It also describes the use of burst packet methods and a modeling approach intended to avoid per-request request-response correlation when deriving latency calculations.

For precision, the blog says ONES computes latency in nanoseconds and models the network analogous to an optical channel. It further notes that the configuration includes parameter flexibility aligned to specific network requirements.

Operational Impact

The blog describes periodic latency calculations as a way to monitor network health, where increases in latency can indicate potential bottlenecks. It also describes regular measurements as a baseline for normal behavior and as data for proactive troubleshooting.

In its use-case framing, the post outlines a “ping-pong mesh” scenario where endpoint-triggered probes traverse the network between endpoints at a set interval to enable ongoing latency checks and assist with resource allocation and performance optimization.

Scalability and Robustness

The blog attributes scalability and robustness to ONES design choices in the latency measurement workflow. It cites validations that include latency measurement under load and the ability to add new nodes for calculations.

It also describes support for a significant number of probes handled by the agent, built-in fault tolerance, optimized resource utilization, and consistent operational and longevity behavior.

Overall, the blog presents ONES network latency measurement backend as an agent-based measurement engine for SONiC and servers that can run periodic ICMP or TCP probes, compute nanosecond-precision latency using burst-based modeling, and support scalable and robust operation for monitoring, SLA checks, and capacity planning. Blog Signals brief is a fact-based summary of the vendor blog.