Predictive Packet Forwarding for Network Switch Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Ethernet packet switches suffer from excessive switch latency, which can lead to system bottlenecks and increased response times when deployed in high-performance parallel computing environments, and they struggle to provide low-latency, policy-rich forwarding for multiplexed traffic types like voice, video, and data without compromising switch performance.
Innovation Solution
The network switch architecture employs flow prediction using signatures to reduce latency by determining a packet's flow prior to completing the lookup process, allowing early transmission of packet bits and optimizing switching speed, and uses shared input and output memories with per-port predictors and caches to enhance prediction accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional lookup-based flow classification is used in Ethernet packet switches, then forwarding accuracy is maintained, but switch latency becomes excessive
Solution Approach 1:
The patent applies preliminary action by performing flow prediction using signatures before the traditional lookup process is complete. The system predicts the flow based on packet signatures and initiates forwarding actions in advance, while the full lookup process continues in parallel to verify the prediction. This allows the switch to prepare forwarding decisions ahead of time, reducing overall latency without sacrificing forwarding accuracy.
Solution Approach 2:
The patent introduces an intermediary prediction mechanism that sits between packet reception and final forwarding decision. The signature-based prediction system acts as an intermediary that provides preliminary flow classification, which is then verified by the traditional lookup process. This intermediary layer enables early forwarding preparation while maintaining the reliability of the complete lookup-based verification process.
2Productivity
If flow prediction is implemented to reduce latency, then switching speed improves, but prediction accuracy must be maintained
Solution Approach 1:
The patent applies partial action by using only the necessary portion of packet data to create signatures for flow prediction, rather than waiting for complete packet reception. The system extracts and processes only the relevant signature fields from incoming packets, enabling faster prediction while maintaining sufficient accuracy. The full packet data is still processed in parallel for verification, so the partial processing does not compromise overall prediction reliability.
3Device complexity
If Ethernet packet switches are deployed in high-performance parallel computing environments, then cost and complexity are reduced, but switch latency creates system bottlenecks
Solution Approach 1:
The patent applies preliminary action by preparing flow predictions and forwarding decisions before packets are fully received or processed. The signature-based prediction system initiates flow classification in advance, allowing the switch to have forwarding paths ready before the complete packet processing pipeline finishes. This reduces the effective latency experienced in high-performance computing environments while maintaining the cost-effectiveness of standard Ethernet switches.
4Adaptability or versatility
If rich forwarding features are provided for multiplexed traffic types, then service quality is improved, but switch performance degrades
Solution Approach 1:
The patent applies segmentation by separating the forwarding process into distinct stages: signature-based flow prediction, traditional lookup verification, and final forwarding execution. This segmentation allows rich forwarding features to be implemented through the flexible signature matching and flow classification mechanisms, while the performance-critical path uses the optimized prediction results. The segmented approach enables versatile traffic handling without degrading overall switch performance.
Data Source
AI summary
A network switch includes a predictor using data in a packet to predict a flow for the packet. A forwarding engine forwards at least a portion of the packet on a switch fabric to an egress port in the switch determined from the predicted flow. The forwarding engine is operable to forward the packet on the switch fabric to the egress port determined from the predicted flow prior to a lookup module determining a flow from a lookup.