Distributed Route Generation in Cluster Server Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high performance computing systems, centralized management nodes are inefficient in generating and updating routes between compute nodes due to processing load, especially when links in the fabric interconnect become non-operational, leading to communication errors and increased time in identifying unreachable nodes.
Innovation Solution
Each compute node in the cluster server generates and maintains its own routes to other nodes using intermediate nodes, distributing the processing load and allowing for quick identification and updating of routes, even in scenarios where links are non-operational, by employing a hardware model to identify reachability in parallel.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If centralized management nodes generate and update routes, then route management is simplified, but processing load on management nodes increases and route update time increases
Solution Approach 1:
The patent divides the centralized route management function into distributed route generation capabilities at each compute node. Instead of one management node handling all route calculations, each compute node independently generates its own routes and receives routes from neighboring nodes, segmenting the processing load across multiple nodes to improve overall route generation speed while maintaining manageable complexity through localized decision-making
Solution Approach 2:
Each compute node performs self-service by autonomously generating its own routes to other nodes in the cluster. The nodes independently calculate routes based on the fabric interconnect topology and share these routes with neighbors, eliminating the need for centralized route management and significantly reducing the processing burden on any single management node while accelerating route deployment
2Stability of the object's composition
If centralized management nodes update routes when links fail, then route consistency is maintained, but time to identify unreachable nodes increases
Solution Approach 1:
The patent implements preliminary action by having each compute node continuously maintain updated routes to all other nodes and proactively share these routes with neighboring nodes before failures occur. When a link fails, the distributed route information already present at each node enables immediate identification of unreachable nodes without waiting for centralized detection, thus maintaining route consistency while dramatically reducing the time to respond to failures
Solution Approach 2:
The system employs feedback mechanisms where compute nodes monitor the operational status of fabric interconnect links and immediately communicate link failure information to neighboring nodes. This distributed feedback loop enables each node to update its route tables in real-time based on actual network conditions, maintaining route consistency across the cluster while rapidly identifying unreachable nodes through localized detection rather than centralized polling
3Productivity
If distributed route generation is implemented, then processing load is reduced, but device complexity at individual nodes increases
Solution Approach 1:
The patent segments the complex route generation task into simpler sub-tasks that can be performed at each compute node. Instead of requiring full centralized control logic at management nodes, each node only needs to generate routes to directly connected neighbors and exchange this information with adjacent nodes. This segmentation of the routing function reduces the processing complexity at individual nodes while collectively achieving high route generation efficiency across the distributed system
Solution Approach 2:
The patent implements universality by designing a standardized route generation and exchange protocol that can be executed by any compute node in the cluster. Each node performs the same multi-functional role of generating local routes, receiving routes from neighbors, validating route information, and updating its route table. This universal approach allows any node to handle route management tasks, distributing the processing load efficiently while keeping individual node complexity manageable through standardized procedures
Data Source
AI summary
Each compute node of a cluster compute server generates and maintains route information for routing messages to other nodes of the server. Each compute node identifies the other nodes connected to a fabric interconnect and generates, based on a set of routing constraints, routes to each of the other nodes. Each compute node communicates messages to other nodes of the server via the generated routes. Because the routes are generated at each compute node the processing load to generate the routes is distributed among the compute nodes.