Quality of Outcome

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶

This Internet-Draft will expire on 4 December 2025.¶

Copyright Notice

Copyright (c) 2025 IETF Trust and the persons identified as the document authors. All rights reserved.¶

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶

Table of Contents

1. Introduction

This document introduces the Quality of Outcome network score. Quality of Outcome is a network quality score designed to be easy to understand, while at the same time being objective, adaptable to different network quality needs, and allowing advanced analysis to identify the root cause of network problems. This document defines a network requirement framework that allows application developers to specify their network requirements, along with a way to create a simple user-facing metric based on comparing application requirements to measurements of network performance. The framework builds on Quality Attenuation [TR-452.1], enabling network operators to achieve fault isolation and effective network planning through composability.¶

Quality attenuation is a network quality metric that meets most of the criteria set out in the requirements; it can capture the probability of a network satisfying application requirements, it is composable, and it can be compared to a variety of application requirements. The part that is yet missing is how to present quality attenuation results to end-users and application developers in an understandable way. A per-application, per application-type, or per-SLA approach is most appropriate here. The challenge lies in specifying how to simplify enough without losing too much in terms of precision and accuracy.¶

Taking a probabilistic approach is key because the network stack and application's network quality adaptation can be highly complex. Applications and the underlying networking protocols make separate optimizations based on their perceived network quality over time and saying something about an outcome with absolute certainty is practically impossible. It is possible, however, to make educated guesses on the probability of outcomes.¶

This document proposes representing network quality as a minimum required throughput and set of latency and loss percentiles. Application developers, regulatory bodies and other interested parties can describe network requirements in the same manner. This document defines a distance measure between perfect and unusable. This distance measure can, with some assumptions, calculate something that can be simplified into statements such as “A Video Conference has a 93% chance of being lag free on this network” all while making it possible to use the framework both for end-to-end test and analysis from within the network.¶

The work proposes a minimum viable framework, and often trades precision for simplicity. The justification for this is to ensure adoption and usability in many different contexts such as active testing from applications and monitoring from network equipment. To counter the loss of precision, is it necessary to combine measurement results with a description of the measurement approach that allow for analysis of the precision.¶

2. Motivation

This section describes the features and attributes a network quality framework must have to be useful for different stakeholders: application developers, end-users, and network operators/vendors. At a high level, end-users need an understandable network metric. Application developers require a network metric that allows them to evaluate how well their application is likely to perform given the measured network performance. Network operators and vendors need a metric that facilitates troubleshooting and optimization of their networks. Existing network quality metrics and frameworks typically address the needs of one or two of these stakeholders, but the authors have yet to find one that bridges the needs of all three.¶

A key motivation for the Quality of Outcome (QoO) framework is to bridge the gap between the technical aspects of network performance and the practical needs of those who depend on it. While solutions exist for many of the problems causing high and unstable latency in the Internet, the incentives to deploy them have remained relatively weak. A unifying framework for assessing network quality, can serve to strengthen these incentives significantly.¶

Bandwidth alone is necessary but not sufficient for high-quality modern network experiences. Idle latency, working latency, jitter, and unmitigated packet loss are major causes of poor application outcomes. The impact of latency is widely recognized in network engineering circles [BITAG], but benchmarking the quality of network transport remains complex. Most end-users are unable to relate to metrics other than Mbps, which they have long been conditioned to think of as the only dimension of network quality.¶

Real Time Response under load tests [RRUL] and Responsiveness [RPM] make significant strides toward creating a network quality metric that is far closer to application outcomes than bandwidth alone. The latter, in particular, is successful at being relatively relatable and understandable to end-users. However, as noted in [RPM], “Our networks remain unresponsive, not from a lack of technical solutions, but rather a lack of awareness of the problem.” This lack of awareness means operators have little incentive to improve network quality beyond increasing throughput. Despite the availability of open-source solutions, vendors rarely implement them. The root cause lies in the absence of a universally accepted network quality framework that captures how well applications are likely to perform.¶

A recent IAB workshop on measuring internet quality for end-users identified a key insight: users care primarily about application performance rather than network performance. Among the conclusions was the statement, "A really meaningful metric for users is whether their application will work properly or fail because of a lack of a network with sufficient characteristics" [RFC9318]. Therefore, one critical requirement of a meaningful framework is its ability to answer the question, "Will an application work properly?"¶

Answering this question requires several considerations. First, the Internet is inherently stochastic from the perspective of any given client, so certainty i unattainable. Second, different applications have varying needs and adapt differently to network conditions. A framework aiming to answer this question must accommodate diverse application requirements. Third, end-users have individual tolerances for degradation in network conditions and the resulting effects on application experience. These variations must be factored into the framework's design.¶

3. Background

The foundation of the framework is Quality Attenuation [TR-452.1]. This work will not go into detail about how to measure Quality Attenuation, but some relevant techniques are:¶

• Active probing with TWAMP Light [RFC5357] / STAMP [RFC8762] / IRTT [IRTT]¶

• Latency Under Load Tests¶

• Speed Tests with latency measures¶

• Simulating real traffic¶

• End-to-end measurements of real traffic¶

• TCP SYN ACK / DNS Lookup RTT Capture¶

• Estimation¶

Quality Attenuation represents quality measurements as distributions. Using Latency distributions to measure network quality is nothing new and has been proposed by various researchers/practitioners [Kelly][RFC8239][RFC6049]. The novelty of the Quality Attenuation metric is to view packet loss as infinite (or too late to be of use e.g. > 3 seconds) latency [TR-452.1].¶

Latency Distributions can be gathered via both passive monitoring and active testing. The active testing can use any type of traffic, such as TCP, UDP, or QUIC. It is OSI Layer and network technology independent, meaning it can be gathered in an end-user application, within some network equipment, or anywhere in between. Passive methods rely on observing and time-stamping packets traversing the network. Examples of this include TCP SYN and SYN/ACK packets and the QUIC spin bit.¶

A key assumption behind the choice of latency distribution is that different applications and application categories fail at different points of the latency distribution. Some applications, such as downloads, have lenient latency requirements when compared to real-time application. Video Conferences are typically sensitive to high 90th percentile latency and to the difference between the 90th and the 99th percentile. Online gaming typically has a low tolerance for high 99th percentile latency. All applications require a minumum level of throughput and a maximum packet loss rate. A network quality metric that aims to generalize network quality must take the latency distribution, throughput, and packet loss into consideration.¶

Two distributions can be composed using convolution [TR-452.1].¶

4. Sampling requirements

To ensure broad applicability across diverse use cases, this framework deliberately avoids prescribing specific conditions for sampling such as fixed time intervals or defined network load levels. This flexibility enables deployment in both controlled and real-world environments.¶

At its core, the framework requires only a latency distribution. When measurements are taken during periods of network load, the result naturally includes latency under load. In scenarios such as passive monitoring of production traffic, capturing artificially loaded conditions may not always be feasible, whereas passively observing the actual network load may be possible.¶

Modeling the full latency distribution may be too complex to allow for easy adoption of the framework, and reporting latency at selected percentiles offers a practical compromise between accuracy and deployment considerations. A commonly accepted set of percentiles spanning from the 0th to the 100th in a logarithmic-like progression has been suggested by others [BITAG] and is recommended here: [0th, 10th, 25th, 50th, 75th, 90th, 95th, 99th, 99.9th, 100th].¶

The framework is agnostic to traffic direction but mandates that measurements specify whether latency is one-way or round-trip.¶

Importantly, the framework does not enforce a minimum sample count. This means that even a small number of samples (e.g., 10) could technically constitute a distribution—though such cases are clearly insufficient for statistical confidence. The intent is to balance rigor with practicality, recognizing that constraints vary across devices, applications, and deployment environments.¶

To support reproducibility and enable confidence analysis, each measurement must be accompanied by the following metadata:¶

• Description of the measurement path¶

• Timestamp of first sample¶

• Total duration of the sampling period¶

• Number of samples collected¶

• Sampling method, including:¶

  • Cyclic: One sample every N milliseconds (specify N)¶

  • Burst: X samples every N milliseconds (specify X and N)¶

  • Passive: Opportunistic sampling of live traffic (non-uniform intervals)¶

These metadata elements are essential for interpreting the precision and reliability of the measurements. As demonstrated in [QoOSimStudy], low sampling frequencies and short measurement durations can lead to misleadingly optimistic or imprecise Quality of Outcome (QoO) scores.¶

5. Describing Network Requirements

This work builds upon the work already proposed in the Broadband Forum standard called Quality of Experience Delivered (QED/TR-452) [TR-452.1], which defines the Quality Attenuation metric. In essence, QoO describes network requirements as a list of percentile and latency requirement tuples. In other words, a network requirement may be expressed as: The network requirement for this app quality level/app/app category/SLA is “at 4 Mbps, 90% of packets needs to arrive within 100 ms, 100% of packets needs to arrive within 200ms”. This list can be as simple as “100% of packets need to arrive within 200ms” or as long as you would like. For the sake of simplicity, the requirements percentiles must match one or more of the percentiles defined in the measurements, i.e., one can set requirements at the [0th, 10th, 25th, 50th, 75th, 90th, 95th, 99th, 99.9th, 100th] percentiles. Packet loss must be reported as a separate value.¶

Applications do of course have throughput requirements, and thus a complete framework for application-level network quality must also take capacity into account. Insufficient bandwidth may give poor application outcomes without necessarily inducing a lot of latency. Therefore, the network requirements should include a minimum throughput requirement. A fully specified requirement can be thought of as specifying the latency and loss requirements to be met while the end-to-end network path is loaded in a way that is at least as demanding of the network as the application itself. This may be achieved by running the actual application and measuring delay and loss alongside it, or by generating artificial traffic to a level at least equivalent to the application traffic load.¶

Whether the requirements are one-way or two-way must be specified. Where the requirement is one-way, the direction (uplink or downlink) must be specified. If two-way, a decomposition into uplink and downlink measurements may be specified.¶

Until now, network requirements and measurements are what is already standardized in the BBF TR-452 (aka QED) framework [TR-452.1]. The novel part of this work is what comes next. A method for going from Network Requirements and Network Measurements to probabilities of good application outcomes.¶

To do that it is necessary to make articulating the network requirements a little bit more complicated. A key design goal was to have a distance measure between perfect and unusable, and have a way of quantifying what is ‘better’.¶

The requirements specification is extended to include the quality required for perfection and a quality threshold beyond which the application is considered unusable.¶

This is named Network Requirements for Perfection (NRP). As an example: At 4 Mbps, 99% of packets need to arrive within 100ms, 99.9% within 200ms (implying that 0.1% packet loss is acceptable) for the outcome to be perfect. Network Requirements Point of Unusability (NRPoU): If 99% of the packets have not arrived after 200ms, or 99.9% within 300ms, the outcome will be unusable.¶

Where the NRPoU percentiles and NRP are a required pair then neither should define a percentile not included in the other - i.e., if the 99.9th percentile is part of the NRPoU then the NRP must also include the 99.9th percentile.¶

6. Creating network requirement specifications

A detailed description of how to create a network requirement specification is out of scope for this document, but this section will provide a rough outline for how it can be achieved. Additional information about this topic can be found in [QoOAppQualityReqs].¶

When searching for an appropriate network requirement description for an application, the goal is to identify the points of perfection and uselessness for the application. This can be thought of as a search process. Run the application across a network connection with adjustable quality. Gradually adjust the network performance while observing the application-level performance. The application performance can be observed manually by the person performing the testing, or using automated methods such as recording video stall duration from within a video player.¶

Establish a baseline under excellent network conditions. Then gradually add delay, packet loss or decrease network capacity until the application no longer performs perfectly. Continue adding network quality attenuation until the application fails completely. The corresponding network quality levels are the points of perfection and unusability.¶

7. Calculating Quality of Outcome (QoO)

The QoO metric calculates the likelihood of application success based on network performance, incorporating both latency and packet loss. There are three key scenarios:¶

• The network meets all the requirements for perfection. Probability of success: 100%.¶

• The network fails one or more criteria at the Point of Unusableness (NRPoU). Probability of success: 0%.¶

• The network performance falls between perfection and unusable. In this case, a QoO score is computed. The QoO score is calculated by taking the worst score derived from latency and packet loss.¶

Latency Component The latency-based QoO score is computed as follows:¶

QoO_latency = min_{i}(min(max((1 - ((ML_i - NRP_i) / (NRPoU_i - NRP_i))) * 100, 0), 100))¶

Where:¶

• ML_i is the Measured Latency at percentile i.¶

• NRP_i is the Network Requirement for Perfection at percentile i.¶

• NRPoU_i is the Network Requirement Point of Unusableness at percentile i.¶

Packet Loss Component Packet loss is considered as a separate, single measurement that applies across the entire traffic sample, not at each percentile. The packet loss score is calculated using a similar interpolation formula, but based on the total measured packet loss (MLoss) and the packet loss thresholds defined in the NRP and NRPoU:¶

QoO_loss = min(max((1 - ((MLoss - NRP_Loss) / (NRPoU_Loss - NRP_Loss))) * 100, 0), 100)¶

Where:¶

• MLoss is the Measured Packet Loss.¶

• NRP_Loss is the acceptable packet loss for perfection.¶

• NRPoU_Loss is the packet loss threshold beyond which the application becomes unusable.¶

Final QoO Calculation The overall QoO score is the minimum of the latency and packet loss scores:¶

QoO = min(QoO_latency, QoO_loss)¶

Example Requirements and Measured Data:¶

• NRP: 4 Mbps {99%, 250 ms, 0.1% loss}, {99.9%, 350 ms, 0.1% loss}¶

• NRPoU: {99%, 400 ms, 1% loss}, {99.9%, 401 ms, 1% loss}¶

• Measured Latency: 99% = 350 ms, 99.9% = 352 ms¶

• Measured Packet Loss: 0.5%¶

• Measured Minimum Bandwidth: 32 Mbps / 28 Mbps¶

Then the QoO is defined:¶

QoO_latency = min( (min(max((1 - (350 ms - 250 ms) / (400 ms - 250 ms)) * 100, 0), 100), (min(max((1 - (352 ms - 350 ms) / (401 ms - 350 ms)) * 100, 0), 100)) ) = min(33.33, 96.08) = 33.33¶

QoO_loss = min(max((1 - (0.5% - 0.1%) / (1% - 0.1%)) * 100, 0), 100) = 55.56¶

Finally, the overall QoO score is:¶

QoO = min(33.33, 55.56) = 33.33¶

In this example, the application has a 33% chance of meeting the quality expectations on this network, considering both latency and packet loss.¶

Implementation Note: Sensitivity to Sampling Accuracy¶

Based on the simulation results in [QoOSimStudy], overly noisy or inaccurate latency samples can artificially inflate worst-case percentiles, thereby driving QoO scores lower than actual network conditions would warrant. Conversely, coarse measurement intervals can miss short-lived spikes entirely, resulting in an inflated QoO. Users of this framework should consider hardware/software measurement jitter, clock offset, or other system-level noise sources when collecting data, and configure sampling frequency and duration to suit their specific needs.¶

8. How to find network requirements

A key advantage of a measurement that spans the range from perfect to unusable, rather than relying on binary (Good/Bad) or other low-resolution (Terrible/Bad/OK/Great/Excellent) metrics, is the flexibility it provides. For example, a chance of lag-free experience below 20% is intuitively undesirable, while a chance above 90% is intuitively favorable—demonstrating that absolute perfection is not required for the QoO metric to be meaningful.¶

However, it remains necessary to define points representing unusableness and perfection. There is no universally strict threshold at which network conditions render an application unusable. For perfection, some applications may have clear definitions, but for others, such as web browsing and gaming, lower latency is always preferable. To assist in establishing requirements, it is recommended that the Network Requirements for Perfection (NRP) be set at the point where further reductions in latency do not result in a perceivable improvement in end-user experience.¶

Someone who wishes to make a network requirement for an application in the simplest possible way, should do something along these lines.¶

• Simulate increasing levels of latency¶

• Observe the application and note the threshold where the application stops working perfectly¶

• Observe the application and note the threshold where the application stops being useful at all¶

Someone who wishes to find sophisticated network requirements might proceed in this way¶

• Set thresholds for acceptable fps, animation fluidity, i/o latency (voice, video, actions), or other metrics capturing outcomes that directly affects the user experience¶

• Create a tool for measuring these user-facing metrics¶

• Simulate varying latency distribution with increasing levels of latency while measuring the user facing metrics.¶

A QoO score at 94 can be communicated as "John's smartphone has a 94% chance of lag-free Video Conferencing", however, this does not mean that at any point of time there is a 6% chance of lag. It means there is a 6% chance of experiencing lag during the entire session/time-period, and the network requirements should be adjusted accordingly.¶

The reason for making the QoO metric for a session is to make it understandable for an end-user, an end-user should not have to relate to the time period the metric is for.¶

9. Insights from Simulation Results

While the QoO framework itself places no strict requirement on sampling patterns or measurement technology, a recent simulation study [QoOSimStudy] examined the metric’s real-world applicability under varying conditions of:¶

1. Sampling Frequency: Slow sampling rates (e.g., <1 Hz) risk missing rare, short-lived latency spikes, resulting in overly optimistic QoO scores.¶

2. Measurement Noise: Measurement errors on the same scale as the thresholds (NRP, NRPoU) can distort high-percentile latencies and cause artificially lower QoO.¶

3. Requirement Specification: Slightly adjusting the latency thresholds or target percentiles can cause significant changes in QoO, especially when the measurement distribution is near a threshold.¶

4. Measurement Duration: Shorter tests with sparse sampling tend to underestimate worst-case behavior for heavy-tailed latency distributions, biasing QoO in a positive direction.¶

In practice, these findings mean:¶

• Calibrate the combination of sampling rate and total measurement period to capture fat-tailed distributions of latency with sufficient accuracy.¶

• Avoid significant measurement noise where possible (e.g., by calibrating time sources, accounting for clock drift).¶

• Thorougly test application requirement thresholds so that the resulting QoO scores accurately reflect application performance.¶

These guidelines are non-normative but reflect empirical evidence on how QoO performs.¶

10. Insights from user testing

A study involving 25 participants tested the Quality of Outcome (QoO) framework in a real-world settings [QoOUserStudy]. Participants used specially equipped routers in their homes for 10 days, providing both network performance data and feedback through pre- and post-trial surveys.¶

Participants found QoO metrics more intuitive and actionable than traditional metrics (e.g., speed tests). QoO directly aligned with their self-reported experiences, increasing trust and engagement.¶

These results provide supporting evidence for QoO's value as a user-focused tool, bridging technical metrics with real-world application performance to enhance end-user satisfaction.¶

11. Known Weaknesses and open questions

A method has been described for simplifying the comparison between application network requirements and quality attenuation measurements. This simplification introduces several artifacts, the significance of which may vary depending on context. The following section discusses some known limitations.¶

Volatile networks - in particular, mobile cellular networks - pose a challenge for network quality prediction, with the level of assurance of the prediction likely to decrease as session duration increases. Historic network conditions for a given cell may help indicate times of network load or reduced transmission power, and their effect on throughput/latency/loss. However: as terminals are mobile, the signal bandwidth available to a given terminal can change by an order of magnitude within seconds due to physical radio factors. These include whether the terminal is at the edge of cell, or undergoing cell handover, the interference and fading from the local environment, and any switch between radio bearers with differing signal bandwidth and transmission-time intervals (e.g. 4G and 5G). This suggests a requirement for measuring quality attenuation to and from an individual terminal, as that can account for the factors described above. How that facility is provisioned onto indiviudal terminals, and how terminal-hosted applications can trigger a quality attenuation query, is an open question.¶

12. Implementation status

Note to RFC Editor: This section MUST be removed before publication of the document.¶

This section records the status of known implementations of the protocol defined by this specification at the time of posting of this Internet-Draft, and is based on a proposal described in [RFC7942]. The description of implementations in this section is intended to assist the IETF in its decision processes in progressing drafts to RFCs. Please note that the listing of any individual implementation here does not imply endorsement by the IETF. Furthermore, no effort has been spent to verify the information presented here that was supplied by IETF contributors. This is not intended as, and must not be construed to be, a catalog of available implementations or their features. Readers are advised to note that other implementations may exist.¶

According to [RFC7942], "this will allow reviewers and working groups to assign due consideration to documents that have the benefit of running code, which may serve as evidence of valuable experimentation and feedback that have made the implemented protocols more mature. It is up to the individual working groups to use this information as they see fit".¶

13. Conventions and Definitions

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

14. Security Considerations

The Quality of Outcome (QoO) framework introduces a method for assessing network quality based on probabilistic outcomes derived from latency, packet loss, and throughput measurements. While the framework itself is primarily analytical and does not define a new protocol, some security considerations arise from its deployment and use.¶

Measurement Integrity and Authenticity¶

QoO relies on accurate and trustworthy measurements of network performance. If an attacker can manipulate these measurements—either by injecting falsified data or tampering with the measurement process—they could distort the resulting QoO scores. This could mislead users, operators, or regulators into making incorrect assessments of network quality.¶

To mitigate this risk:¶

• Measurement agents can authenticate with the systems collecting or analyzing QoO data.¶

• Measurement data can be transmitted over secure channels (e.g., TLS) to ensure confidentiality and integrity.¶

• Digital signatures may be used to verify the authenticity of measurement reports.¶

Risk of Misuse and Gaming¶

As QoO scores may influence regulatory decisions, service-level agreements (SLAs), or user trust, there is a risk that network operators or application developers might attempt to "game" the system. For example, they might optimize performance only for known test conditions or falsify requirement thresholds to inflate QoO scores.¶

Mitigations include:¶

• Independent verification of application requirements and measurement methodologies.¶

• Use of randomized or blind testing procedures.¶

• Transparency in how QoO scores are derived and what assumptions are made.¶

Privacy Considerations¶

QoO measurements may involve collecting detailed performance data from end-user devices or applications. Depending on the deployment model, this could include metadata such as IP addresses, timestamps, or application usage patterns.¶

To protect user privacy:¶

• Data collection should follow the principle of data minimization, only collecting what is strictly necessary.¶

• Personally identifiable information (PII) should be anonymized or pseudonymized where possible.¶

• Users should be informed about what data is collected and how it is used, in accordance with applicable privacy regulations (e.g., GDPR).¶

Denial of Service (DoS) Risks¶

Active measurement techniques used to gather QoO data (e.g., TWAMP, STAMP, synthetic traffic generation) can place additional load on the network. If not properly rate-limited, this could inadvertently degrade service or be exploited by malicious actors to launch DoS attacks.¶

Recommendations:¶

• Implement rate limiting and access control for active measurement tools.¶

• Ensure that measurement traffic does not interfere with critical services.¶

• Monitor for abnormal measurement patterns that may indicate abuse.¶

Trust in Application Requirements¶

QoO depends on application developers to define Network Requirements for Perfection (NRP) and Network Requirements Point of Unusableness (NRPoU). If these are defined inaccurately—either unintentionally or maliciously—the resulting QoO scores may be misleading.¶

To address this:¶

• Encourage peer review and publication of application requirement profiles.¶

• Where QoO is used for regulatory or SLA enforcement, require independent validation of requirement definitions.¶

15. IANA Considerations

This document has no IANA actions.¶

16. References

Acknowledgments

The authors would like to thank Will Hawkins, Stuart Cheshire, Jason Livingood, Olav Nedrelid, Greg Mirsky, Tommy Pauly, Marcus Ihlar, Tal Mizrahi, Ruediger Geib, Mehmet Şükrü Kuran and Neil Davies for their feedback and input to this document.¶

Authors' Addresses