Backward Congestion Notification for Network Buffer Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-speed networks with small buffer sizes and low latency face challenges in managing network congestion effectively, as increasing buffer sizes proportionally is economically infeasible, leading to rapid buffer filling during congestion.
Innovation Solution
Implementing a network interface system with logic devices that determine frame addresses, calculate flow hashes, and insert congestion management tags into frames to manage congestion by sending feedback messages from congestion points to reaction points, which use rate limiters to mitigate congestion based on received notifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If buffer sizes are increased to manage network congestion, then congestion management capability is improved, but cost and device complexity increase proportionally
Solution Approach 1:
The patent segments the network into core network points (congestion points) and edge devices (reaction points). Congestion management functionality is segmented and pushed from the core to the edge, where rate limiters are implemented at reaction points instead of requiring large buffers at core network devices. This segmentation allows effective congestion management without proportionally increasing buffer sizes at central points.
Solution Approach 2:
The patent inverts the traditional congestion management approach by pushing congestion handling capabilities from the core network to the network edge. Instead of managing congestion centrally with large buffers, the system implements rate limiters at edge devices that actively prevent congestion before it reaches the core, effectively managing congestion with smaller buffers at central points.
2Loss of time
If buffer sizes are kept small to maintain low latency, then latency is reduced, but congestion management capability deteriorates
Solution Approach 1:
The patent implements preliminary action by placing rate limiters at reaction points (edge devices) that proactively regulate traffic flows before they can cause congestion in the core network. This preventive approach allows small buffers to be maintained while congestion management capability is preserved, as traffic is controlled before it reaches bottleneck points.
Solution Approach 2:
The patent implements a feedback mechanism where congestion points send backward congestion notifications (BCN messages) to reaction points when congestion is detected. This feedback loop enables dynamic adjustment of rate limiting at the edge based on actual congestion conditions, maintaining effective congestion management with small buffers and low latency.
3Reliability
If rate limiters are implemented at reaction points to mitigate congestion, then congestion control is improved, but device complexity at edge devices increases
Solution Approach 1:
The patent extracts the congestion management functionality from core network devices and places it at edge devices (reaction points). By taking out the rate limiting function and positioning it at the network edge, the core network devices can remain simpler while edge devices handle the complex congestion control logic, distributing complexity to where it is most effective.
Data Source
AI summary
In one embodiment, an apparatus comprises a network interface system having at least one input port configured for receiving frames and a logic system comprising at least one logic device. The logic system may be configured to perform the following functions: determining a source address and a destination address of a frame received at an ingress port; calculating a flow hash based at least upon the source address and the destination address; forming a congestion management (“CM”) tag that includes the flow hash; inserting the CM tag in the frame; and forwarding the frame to the destination address.