Distributed Buffer Memory for Network Congestion Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data center networks face challenges with congestion management, particularly in incasting scenarios where multiple ingress ports send excessive traffic to a single egress port, leading to inefficiencies and increased costs due to the need for high-capacity memory and complex control mechanisms.
Innovation Solution
Implementing a distributed buffer memory system where packets are intermediately directed to dedicated buffer devices across related nodes in the network, allowing for efficient load distribution and congestion management without the need for extensive high-capacity memory or complex control messaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If packets are buffered at a centralized location (ingress or egress), then congestion can be managed, but memory capacity and complexity increase significantly
Solution Approach 1:
The patent segments the centralized buffer into multiple distributed buffers located at different nodes in the network. Each node maintains its own buffer, dividing the total buffering capacity across multiple locations. This segmentation reduces the memory capacity requirement at any single location and distributes control complexity, making the system more manageable and scalable.
Solution Approach 2:
The patent introduces a new dimension to buffering by utilizing the network topology itself as a distributed buffer structure. Instead of adding buffering capacity in the traditional vertical dimension (more memory at one location), the solution distributes buffering horizontally across multiple network nodes, effectively using the network's spatial dimension to manage congestion.
2Reliability
If high-capacity memory is deployed at each ingress port to handle local congestion, then incasting scenarios are managed, but cost and power consumption increase
Solution Approach 1:
The patent merges the buffering functions of multiple ingress ports into a shared distributed buffer pool across the network. Instead of each ingress port maintaining separate high-capacity memory, the system combines buffering resources network-wide, allowing statistical multiplexing and reducing total memory capacity requirements while maintaining incasting scenario handling capability.
3Productivity
If complex control messaging is implemented to manage distributed buffers, then network performance is optimized, but system complexity increases
Solution Approach 1:
The patent implements self-service mechanisms where network nodes autonomously make buffering decisions based on local conditions and simple signaling protocols. Each node independently manages its buffer space, allocating and deallocating resources without requiring complex centralized control messaging. This self-service approach optimizes network performance while keeping control mechanisms simple and scalable.
Data Source
AI summary
Systems, methods, and computer-readable media are provided for controlling the distribution of data packets to intermediate buffers, such as Virtual Output Queues (VOQs) of a group of nodes. A method, according to one implementation, includes a step of detecting traffic congestion metrics associated with an egress port of a destination node grouped in a set of related nodes in a network. Each of the related nodes includes a dedicated buffer device. Based on the detected traffic congestion metrics, the method further includes the step of distributing data packets, which are intended for transmission to the egress port of the destination node, intermediately to the dedicated buffer device of one or more of the related nodes.