Network Packet Tagging for Cut-Through Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern high-speed switching systems face delays in end-to-end delivery of network packets due to the need for store and forward buffering, which arises from out-of-order and interleaved completion data from PCIe read requests.
Innovation Solution
The solution involves tagging each cell with a unique packet identifier and byte offset, allowing the receiving device to perform on-the-fly assembly of cells into packets, eliminating the need for store and forward buffering and achieving pure cut-through delivery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If store and forward buffering is used to handle out-of-order completion data, then data integrity is maintained, but end-to-end latency increases
Solution Approach 1:
The patent segments the network packet into multiple fabric cells, each tagged with sequence information. This allows the receiving device to reassemble packets from out-of-order cells without requiring store-and-forward buffering, thus maintaining data integrity while reducing latency.
Solution Approach 2:
The transmitting device performs preliminary tagging of fabric cells with packet identifiers and sequence numbers before injection into the network fabric. This preliminary action enables the receiving device to immediately reassemble packets without buffering delays.
2Loss of time
If pure cut-through delivery is implemented, then end-to-end latency is reduced, but handling of out-of-order cells becomes complex
Solution Approach 1:
The patent introduces intermediary tagging information (packet identifiers and sequence numbers) attached to each fabric cell. This intermediary data structure simplifies the reassembly process at the receiving device, making pure cut-through delivery manageable despite the complexity of handling out-of-order cells.
Solution Approach 2:
The patent changes the parameter representation by adding metadata tags to fabric cells, transforming the problem from complex reassembly logic to simple tag-based sorting and reassembly, thereby reducing device complexity while enabling pure cut-through delivery.
3Productivity
If multiple read requests are submitted in parallel, then throughput is improved, but completion data arrives out-of-order
Solution Approach 1:
The patent implements a feedback mechanism where fabric cells carry sequence number information that enables the receiving device to detect and correct out-of-order arrivals. This feedback loop allows parallel read requests to maintain high throughput while ensuring proper reassembly order.
Solution Approach 2:
By segmenting the data flow into individually tagged fabric cells, the system can handle parallel completion data streams independently and reassemble them in the correct order, thus maintaining both high throughput from parallel requests and proper data composition.
Data Source
AI summary
A method and devices for reducing the delay in end-to-end delivery of network packets may be achieved by having the transmission (TX) side of the device, tag each cell with a unique packet identifier and with a byte offset parameter where the tagging allows the reception (RX) side of the destination device to perform on-the-fly assembly of cells into packets by directly placing them at corresponding host buffer, and the method may be done for multiple packets concurrently, and hence store and forward buffering is not needed in either the source or the destination devices and the lowest possible end-to-end cut-through latency is achieved.