DiffServ Admission Control Using Exponential Moment Delay Bounds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing DiffServ networks face challenges in providing deterministic and statistical delay bounds for differentiated service classes, particularly in ensuring quality of service (QoS) without resorting to separate queues for individual flows, and in handling aggregate flows with varying traffic patterns and burstiness.
Innovation Solution
The solution involves a method for admission control that calculates delay bounds by considering the burstiness and queuing delays of individual traffic streams, using exponential moments for on-off distributions to account for fluctuations, and introducing thresholds for simplified on-line admission decisions, allowing for accurate service quality predictions while optimizing network resource use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate queues for individual flows are used to provide QoS guarantees, then service quality is improved, but device complexity and resource consumption increase
Solution Approach 1:
The patent merges multiple individual flow queues into a single shared queue for aggregate flow treatment. Instead of maintaining separate queues for each flow as in IntServ, the system uses a unified queuing structure that handles all flows collectively, thereby reducing device complexity while still providing QoS guarantees through admission control and traffic shaping mechanisms
2Measurement precision
If detailed occupancy bound calculations are performed for each service class, then delay bound accuracy is improved, but calculation time increases
Solution Approach 1:
The patent performs detailed occupancy bound calculations in advance during network configuration or reconfiguration phases, rather than performing them in real-time during admission control decisions. The results of these preliminary calculations are stored and used to establish simplified occupancy thresholds that can be quickly applied during actual admission control operations, thus maintaining accuracy while reducing real-time computational burden
Solution Approach 2:
The patent replaces complex, time-consuming detailed occupancy bound calculations with simpler, faster-to-compute occupancy thresholds during admission control. These simplified thresholds are sufficient for real-time decision-making and can be quickly updated or recomputed when needed, providing a practical balance between accuracy and computational efficiency
3Device complexity
If aggregate flow treatment is used to reduce per-flow state, then device complexity is reduced, but delay bound accuracy for time-fragile traffic decreases
Solution Approach 1:
The patent applies different admission control strategies and occupancy thresholds tailored to specific service classes and traffic types within the aggregate flow framework. By considering local characteristics of different traffic streams (such as burstiness parameters and service class requirements), the system maintains accurate delay bounds for time-fragile traffic while still using aggregate-level queuing structures, thus resolving the contradiction between simplicity and precision
Data Source
AI summary
The present invention relates to on-line admission-control decisions. Specifically, the invention concerns general delay bounds for both deterministic and statistical cases for Differentiated Services (DiffServ) networks. More specifically, a detailed method of calculation in each case is followed by simpler methods of calculation that are more appropriate for on-line admission-control decisions. Relatively involved occupancy bound calculations for various service classes take place only at the time of network configuration or reconfiguration. At the time of admission control only simple occupancy threshold compliance calculations need to be performed. Concrete illustrations are provided for deriving bounds for the EF and AF classes provided by DiffServ. These results are applicable to both layer-3 networks that support DiffServ and layer-2 networks that support the more restricted class of service functions.