We frequently get this question from customers and partners, so we decided to write a short blog post to address it. It’s a topic that seems straightforward at first — but once you scratch the surface, it turns out to be more complex than it appears.
At first glance, TWAMP Full might look like a practical way to measure one-way packet loss. However, in real-world deployments, relying solely on TWAMP Full can result in misleading or incomplete data — particularly when loss directionality is important. Here’s why.
Limitations in the Standard
Several sections of RFC 5357, that defines TWAMP, limit the protocol’s ability to accurately detect one-way loss:
- Section 3.9 states that measurement methods do not require packet-level data to be collected by the Session-Reflector. In practice, this means the Session-Reflector is a stateless engine — it doesn’t keep any information about the packets it receives or sends, nor can it expose such information to the Session-Sender.
- Section 4.2.1 (paragraph 2) does not require the Session-Reflector to maintain a sequence number per TWAMP test session.
- RFC Errata 1590 later clarified this ambiguity, stating that the sequence number should be maintained per TWAMP test session — but since this is not part of the original standard, it is not universally implemented.
Why This Matters in Production
Even with RFC Errata 1590 in place, several challenges still persist in real-world environments.
First, Errata 1590 is not implemented consistently across vendors. Some devices correctly maintain sequence numbers per TWAMP session, while others do not. In some implementations, the Session-Reflector simply copies the Sender’s sequence number instead of generating its own. Without proper session-based sequence tracking, it becomes impossible to determine the direction of packet loss.
Second, one-way loss remains ambiguous in real-world edge cases. For example, if the final measurement packet or packets are lost, there is no way to tell whether the loss occurred during transmission or reception. Imagine sending 100 packets and receiving only 95 — with the last 5 missing. In such cases, there’s no reliable method to determine whether the packets were lost on the outbound or return path.
How 5×9 LMS Address This
To overcome the limitations of standard TWAMP for one-way loss detection, 5×9 LMS introduces an intelligent extension to TWAMP functionality and communication between Probes. Here’s how our approach works:
- We’ve developed a stateful-capable Session-Reflector, powered by a 5×9 engine that tracks packet-level statistics per session
- We’ve also built an intelligent Session-Sender, which can automatically switch to stateful mode when two-way packet loss is detected. This on-demand behavior ensures that stateful mode is only used when necessary — helping us stay lightweight and efficient. 😉
- When operating in stateful mode, the Session-Sender retrieves packet-level data from the Session-Reflector. This enables 100% accurate detection of one-way loss and its direction, even in challenging scenarios such as final-packet(s) loss.
The Best Parts
The 5×9 extension is fully backward compatible with standard TWAMP deployments. It works seamlessly on both the Sender and Reflector sides, ensuring smooth interoperability even in mixed environments. If the remote TWAMP measurement endpoint isn’t an intelligent 5×9 Probe, or if one-way loss information isn’t required, the extension simply isn’t used.
Moreover, the solution operates entirely in TWAMP Light continuous mode, with full support for multiple QoS classes. This approach eliminates the complexity and signaling overhead associated with TWAMP Full sessions, resulting in a lighter, faster and more scalable implementation.
Conclusion
TWAMP Full, while perfectly suitable for two-way measurements, has inherent limitations when it comes to one-way loss detection — particularly when vendor implementations vary or precise directional insights are required.
5×9’s enhanced approach bridges these gaps by extending standard TWAMP with intelligent, on-demand functionality. The result is a solution that delivers the measurement clarity, directional accuracy and operational reliability modern networks require.
Author: Mario Jurcevic, 5×9 Networks
