Switch Control Packets for Manager-Free In-Network Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The deployment and maintenance of in-network computing managers in collective communication systems are complex, especially in large-scale networks, hindering efficient performance optimization.
Innovation Solution
A method for processing control packets in a collective communication system that allows direct querying of in-network computing capabilities through multiplexing the system context, enabling efficient INC offloading without the need for a control plane manager or computing node daemon, and supporting existing network protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an independent manager with subnet manager and aggregation manager is deployed for in-network computing, then in-network computing capability is achieved, but deployment complexity and maintenance difficulty increase significantly
Solution Approach 1:
The patent extracts the in-network computing capability from the complex manager architecture and embeds it directly into the switch hardware. The switch integrates the forwarding capability and computing capability, eliminating the need for separate subnet managers and aggregation managers. This extraction simplifies the overall system architecture while maintaining the in-network computing function.
Solution Approach 2:
The switch is designed to perform multiple functions: it simultaneously handles packet forwarding, in-network computing operations, and resource allocation. By making the switch multi-functional, the patent eliminates the need for separate dedicated managers, thereby reducing deployment complexity while maintaining comprehensive in-network computing capability.
2Productivity
If an independent manager is deployed for in-network computing, then computing offloading is enabled, but maintenance difficulty increases especially in large-scale networks
Solution Approach 1:
The patent removes the manager components from the architecture and embeds the computing offloading functionality directly into the switch. This extraction of management functions from separate software components and integration into hardware simplifies maintenance, especially in large-scale networks where manager coordination becomes increasingly complex.
Solution Approach 2:
The switch performs self-configuration and self-management for in-network computing operations. It automatically allocates resources, manages computing tasks, and handles packet forwarding without requiring external manager intervention. This self-service capability significantly reduces maintenance difficulty while maintaining full computing offloading functionality.
3Reliability
If control plane manager and computing node daemon are used, then in-network computing is coordinated, but dependency complexity increases
Solution Approach 1:
The patent extracts the coordination functionality from the control plane manager and computing node daemon architecture and implements it directly within the switch. This eliminates the dependencies on external management software components while maintaining reliable coordination of in-network computing operations.
Solution Approach 2:
The switch acts as an intermediary that directly performs in-network computing operations without requiring coordination from separate control plane managers or computing node daemons. This intermediary role eliminates the complex inter-component dependencies while maintaining reliable operation.
Data Source
AI summary
The present disclosure relates to methods and apparatuses for processing a control packet in a collective communication system, where the collective communication system includes a switch network and multiple computing nodes, and the switch network includes a first switch. In one example method, the first switch forwards a query packet transmitted by a source node to a destination node, where the query packet is generated by the source node based on a context of the collective communication system. Then, the first switch forwards a notification packet transmitted by the destination node to the source node, where the notification packet carries an in-network computing capability of the switch network.