Abstract
Proportional Delay Differentiation (PDD) is a critical Quality of Service (QoS) mechanism designed to maintain fixed delay ratios between traffic classes. Traditional models, such as BPR and JoBS, rely on heavy-load assumptions that fail in non-saturated network conditions where idle periods occur. This paper introduces a fuzzy-logic, rule-based rate-allocation approach that dynamically adjusts service rates to maintain consistent PDD. Experimental results demonstrate that the rule-based controller maintains target delay ratios at loads as low as 70%, significantly outperforming existing linear feedback and mathematical models.
1. Introduction
The goal of Proportional Delay Differentiation (PDD) is to ensure that the queueing delay of a lower-priority traffic class is a fixed multiple of a higher-priority class, independent of total traffic volume. While existing approaches like BPR (Base-Priority-Ratio) or JoBS (Joint Buffer Management and Scheduling) work well under heavy load, they struggle in "non-heavily loaded" systems. In these scenarios, frequent idle periods—intervals where the queue is empty—distort the assumed inverse relationship between service rate and delay. This paper addresses the need for a more robust control mechanism that can handle the non-linearities introduced by varying network loads.
2. Methodology: Fuzzy Logic Control
The authors propose a feedback control loop grounded in fuzzy logic to approximate the complex, non-linear relationship between service rates and achieved delay ratios.
2.1 The Control Loop
The system monitors two primary inputs: the error ($e(k)$), defined as the difference between the target and actual delay ratios, and the change in error ($\Delta e(k)$). These inputs are processed through a fuzzy logic controller to determine the necessary adjustment to the service rate ($u(k)$).
2.2 Fuzzy Inference Mechanism
The controller operates in three stages:
- Fuzzification: Numerical error values are mapped to linguistic categories such as "Positive Large" or "Negative Small."
- Inference Mechanism: A set of heuristic "If-Then" rules is applied. For example, if the error is positive and increasing, the controller aggressively adjusts the service rate to reverse the trend.
- Defuzzification: The linguistic conclusions are converted back into a precise numerical adjustment ($\Delta u(k)$) for the next sampling period.
3. Experimental Setup
The proposed approach was evaluated using a controlled network simulation with the following parameters:
- Traffic Profile: Two classes of "bursty" traffic following a Pareto distribution (shape parameter 1.5).
- Packet Mix: A realistic distribution of packet sizes (40B, 550B, and 1500B).
- Load Conditions: System load was varied from 60% to 95% to test robustness across different congestion levels.
4. Results and Discussion
The rule-based approach was compared against JoBS, BPR, and Linear Feedback Control (LFB).
4.1 Performance Consistency
The rule-based controller demonstrated superior consistency, maintaining the target delay ratio of 4.0 starting from a 70% load. In contrast, JoBS and BPR required at least 90% load to achieve the target ratio. Linear Feedback Control failed to provide stable PDD services in most test cases involving mixed busy and idle periods.
4.2 Handling the Conservation Law
At very low loads (60%), the "Conservation Law" of networking makes certain delay ratios physically impossible to achieve in a work-conserving system. While no method could hit the target at this load, the rule-based approach stayed closest to the theoretical maximum feasible ratio, whereas other methods showed significant degradation.
5. Conclusion
By utilizing fuzzy logic to bypass the need for an exact analytical model of idle periods, the proposed rule-based approach provides a more reliable solution for PDD in real-world network conditions. Future work will focus on scaling this approach to a larger number of traffic classes and evaluating the computational overhead on high-speed routing hardware.
Summary updated