Switch Routing Table Segmentation for Supercomputer Memory Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current routing solutions for supercomputers are limited by memory space constraints and high calculation times, making it difficult to cover all possible routes in complex network topologies, especially in high-performance computing environments where vast amounts of data need to be efficiently routed.
Innovation Solution
A method and system for adaptive routing that pre-records a list of directly interfaced switches and uses aggregated information to select the optimal output port based on characteristics like hop count and congestion, allowing data to be routed efficiently through a network by identifying the best path to a leaf switch serving multiple destinations, thereby reducing memory usage and calculation complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional routing tables are used to cover all possible paths in complex network topologies, then routing completeness is improved, but memory space requirements increase significantly
Solution Approach 1:
The patent segments the routing table into two distinct parts: a routing information base (RIB) that stores aggregated routing information with destination prefixes and next-hop switches, and a forwarding information base (FIB) that stores pre-calculated shortest path information. This segmentation allows the system to maintain routing completeness while reducing memory usage by separating comprehensive routing data from frequently accessed forwarding data.
Solution Approach 2:
The patent implements preliminary action by pre-calculating and storing shortest path information in the FIB during an initialization phase or periodic updates. This pre-computation eliminates the need for real-time path calculations during data forwarding, reducing both memory requirements for storing all possible paths and calculation time during operation. The system performs the computationally intensive path-finding work in advance when resources are available.
2Adaptability or versatility
If all possible paths are pre-calculated and stored in switch memory, then routing adaptability is improved, but calculation time increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating shortest paths using algorithms like Dijkstra's during initialization or periodic updates, storing results in the FIB. This eliminates real-time calculation during data forwarding, reducing operational calculation time to simple table lookups while maintaining routing adaptability through periodic updates of the pre-computed paths.
Solution Approach 2:
The patent implements dynamics by allowing the routing system to adapt between static pre-computed paths in the FIB for fast forwarding and dynamic path recalculation when network conditions change. The system can periodically update the FIB with new shortest paths or recalculate routes in response to link failures or topology changes, balancing pre-computation benefits with adaptive response to network conditions.
3Quantity of substance
If hierarchical routing solutions are used to reduce memory usage, then memory efficiency is improved, but routing precision decreases
Solution Approach 1:
The patent segments routing functionality into two precision levels: the RIB stores aggregated routing information with destination prefixes providing broad routing guidance, while the FIB stores precise shortest path information for specific destinations. This segmentation maintains routing precision by ensuring that the FIB contains accurate next-hop information for each destination, while the RIB provides the hierarchical aggregation for memory efficiency.
Solution Approach 2:
The patent creates a composite routing system combining two different data structures with complementary properties: the RIB uses aggregated prefix-based routing information for memory efficiency, while the FIB uses detailed shortest path tables for routing precision. Together, these composite structures achieve both memory efficiency and routing precision that neither structure could achieve alone.
4Adaptability or versatility
If comprehensive routing tables are maintained in each switch, then routing versatility is improved, but device complexity increases
Solution Approach 1:
The patent reduces switch complexity by segmenting the routing system into two simplified components: the RIB that performs aggregated routing lookups and the FIB that performs fast forwarding decisions. Each component has a specialized, simplified data structure and lookup process, avoiding the complexity of maintaining a single comprehensive routing table that supports both hierarchical and precise routing functions.
Data Source
AI summary
Method of routing data in a network, this data originating from a first node and being destined for a second node, this method comprising the following steps: —pre-recording in each of the switches of a list of switches directly interfaced with one or more nodes; —reception of the data by a first switch and determination of its destination; when the first switch is directly interfaced with the second node via an output port-selection then in routing of the data by said output port; otherwise —identification of the switch to which the second node is directly interfaced; —routing of the data to the identified switch via an output port of the first switch, so that the data can be forwarded to said identified switch.