Predictive Queue Depth for Network Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face congestion issues due to rapid increases in network traffic, leading to packet drops and degraded application performance, as existing congestion notification methods may not react quickly enough to mitigate incast conditions.
Innovation Solution
A network interface device predicts queue occupancy levels by tracking the rate of change in buffer occupancy and sends congestion notifications to senders before queue overflow, allowing for proactive reduction in transmission rates or pausing of transmissions to alleviate congestion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional congestion notification methods (ECN/RED) are used, then senders can detect congestion, but the reaction is too late to prevent queue overflow and packet loss
Solution Approach 1:
The patent applies preliminary action by predicting future queue occupancy levels based on current traffic patterns and queue depth trends. Instead of waiting for congestion to occur or for traditional ECN/RED mechanisms to trigger, the system proactively identifies incast conditions before they cause queue overflow, allowing senders to reduce transmission rates in advance and prevent packet loss.
2Measurement precision
If queue occupancy thresholds are monitored continuously, then congestion can be detected, but packet loss occurs before notification reaches senders
Solution Approach 1:
The system performs preliminary prediction of queue occupancy levels using current traffic patterns and queue depth trends. By calculating projected future queue states, the patent enables congestion notification to be sent before actual queue overflow occurs, maintaining both monitoring accuracy and packet delivery efficiency.
Solution Approach 2:
The patent implements a feedback mechanism where the network device continuously monitors queue depth and traffic patterns, predicts future congestion states, and sends notifications back to senders. This closed-loop feedback system allows dynamic adjustment of transmission rates based on predicted queue conditions, improving both measurement precision and productivity.
3Reliability
If congestion notifications are sent after queue overflow, then packet loss has already occurred, but sending notifications earlier requires predicting future queue states
Solution Approach 1:
The patent applies preliminary action by predicting future queue occupancy levels based on current traffic patterns and queue depth trends. Instead of waiting for congestion to occur or for traditional ECN/RED mechanisms to trigger, the system proactively identifies incast conditions before they cause queue overflow, allowing senders to reduce transmission rates in advance and prevent packet loss.
Solution Approach 2:
The system changes the parameter being monitored from static queue depth to dynamic predicted queue occupancy. By tracking trends in queue depth over time and projecting future states, the patent transforms the congestion detection mechanism into a predictive model that balances notification accuracy with manageable computational complexity.
Data Source
AI summary
Examples described herein relate to an apparatus that includes a network interface device comprising circuitry to identify at least one congested queue, predict occupancy level of the at least one congested queue when at least one sender is predicted to receive at least one congestion notification and transmit the at least one congestion notification to the at least one sender through zero or more intermediate nodes. In some examples, to identify at least one congested queue, the circuitry is to identify the at least one congested queue based on at least one fill level. In some examples, to identify at least one congested queue, the circuitry is to identify the at least one congested queue based on at least one predicted fill level at a predicted time the at least one sender receives the at least one congestion notification.