Network Scheduling Weights for Heavy-Tailed Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In modern communication networks, existing scheduling algorithms like MWS struggle to manage a mix of heavy-tailed (HT) and light-tailed (LT) traffic flows effectively, leading to unbounded average delays for LT traffic when both types coexist, as HT flows dominate queue service due to longer backlog rates.
Innovation Solution
The proposed Delay-Based Maximum Power-Weight Scheduling (DMPWS) scheme classifies flows into HT and LT classes based on packet inter-arrival times and HoL delays, assigning higher power factors to LT flows to ensure fair service, using discrete wavelet transforms and extreme value theory for accurate classification, and dynamically adjusts scheduling policies based on traffic conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing scheduling algorithms like MWS are used to manage mixed HT and LT traffic flows, then HT flows receive adequate service due to their longer backlog rates, but LT traffic experiences unbounded average delays because HT flows dominate queue service
Solution Approach 1:
The patent applies local quality by assigning different power factors to different traffic classes (HT and LT) based on their specific characteristics. LT flows receive higher power factors during periods when they are under-served, while HT flows receive appropriate weighting when they need service. This differentiated treatment resolves the contradiction by ensuring each traffic class receives appropriate service quality without one dominating the other.
Solution Approach 2:
The patent implements dynamic scheduling weights that adapt over time based on observed traffic patterns and delay conditions. The power factors are not static but are adjusted dynamically to respond to changing network conditions, allowing the system to prevent LT flow starvation when HT flows dominate and to maintain efficiency when LT flows are active.
2Loss of time
If scheduling weights are increased for LT flows to ensure fair service, then LT traffic delay is reduced, but overall network throughput may decrease due to reduced service to HT flows
Solution Approach 1:
The patent uses dynamic power factors that are adjusted based on current network conditions and traffic class performance. When LT flows experience excessive delay, their power factor is increased to improve service. When HT flows are the primary traffic or network conditions favor HT service, the power factors are adjusted accordingly. This dynamic adjustment ensures throughput is maintained while preventing LT flow starvation.
Solution Approach 2:
The patent changes the scheduling parameter (power factor) based on traffic class and network conditions. By modifying this key parameter dynamically, the system can shift the balance between LT and HT flow service to optimize both delay performance and throughput, rather than using fixed weights that would compromise one or the other.
3Measurement precision
If traffic classification is performed using discrete wavelet transforms and extreme value theory, then accurate classification of HT and LT flows is achieved, but computational complexity and processing overhead increase
Solution Approach 1:
The patent performs traffic classification using wavelet transforms and extreme value theory analysis in advance, before scheduling decisions are made. By pre-analyzing traffic patterns and classifying flows into HT or LT categories upfront, the system achieves accurate classification without repeating complex computations during each scheduling interval, thus reducing real-time processing overhead.
Solution Approach 2:
The patent creates simplified representations of traffic patterns through wavelet analysis and extreme value theory, capturing the essential characteristics of HT and LT flows. These simplified models allow for accurate classification without requiring the full complexity of the original traffic data during scheduling decisions, effectively copying only the necessary features for classification.
Data Source
AI summary
A network element (NE) comprising a receiver configured to receive packets from a plurality of flows, and a processor coupled to the receiver and configured to perform classification on the plurality of flows according to arrivals of the packets to classify the plurality of flows into a heavy-tailed (HT) class and a light-tailed (LT) class assign scheduling weights to the plurality of flows according to the classification, and select a scheduling policy to forward the packets of the plurality of flows according to the scheduling weights.