Metabase-Aided Fat-Tree Network Reconfiguration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The reconfiguration of interconnection networks, such as InfiniBand networks, often requires costly and extensive modifications due to configuration-oblivious routing, which affects performance and efficiency, especially in large-scale high-performance computing systems like Exascale computing, where frequent reconfigurations are necessary to maintain connectivity and performance.
Innovation Solution
The implementation of a method that calculates and stores leaf-switch to leaf-switch multipaths using a master subnet manager, allowing for minimal routing updates and metabase-aided network reconfiguration, which preserves existing routes and reduces reconfiguration overhead by using a two-phase routing technique that separates path calculation and allocation, enabling efficient reconfiguration without significant performance penalties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If configuration-oblivious routing is used for network reconfiguration, then connectivity can be restored, but the number of route modifications increases substantially and reconfiguration becomes costly
Solution Approach 1:
The patent pre-calculates and stores multiple alternative paths between leaf-switches in a metabase before reconfiguration is needed. When reconfiguration is triggered, the system can immediately select from pre-computed paths without performing extensive real-time calculations, thereby reducing the number of route modifications required while maintaining connectivity.
Solution Approach 2:
The system dynamically adapts the routing configuration by selecting from multiple pre-calculated paths based on current network conditions and failure scenarios. The metabase stores various path options that can be flexibly chosen during reconfiguration, allowing the system to minimize route modifications by selecting paths that require the least changes from the current configuration.
2Reliability
If extensive route modifications are performed during reconfiguration, then connectivity is maintained, but performance degrades due to reconfiguration overhead
Solution Approach 1:
By pre-calculating multiple alternative paths and storing them in the metabase before reconfiguration events, the system eliminates the need for extensive real-time path computation during reconfiguration. This preliminary action reduces reconfiguration overhead and minimizes performance degradation, as the system can quickly select from pre-computed paths rather than calculating new routes from scratch.
Solution Approach 2:
The system skips the time-consuming step of real-time path calculation during reconfiguration by utilizing pre-computed paths stored in the metabase. This allows the reconfiguration process to proceed rapidly, minimizing the time the network spends in a transitional state and reducing performance impact.
3Reliability
If frequent reconfigurations are performed to maintain connectivity in large systems, then reliability is improved, but the cost and time of reconfiguration increases
Solution Approach 1:
The system performs path calculations in advance and stores multiple alternative paths in the metabase, so that when reconfiguration is needed, the system can immediately select from pre-computed options without time-consuming real-time calculations. This preliminary preparation significantly reduces reconfiguration time, enabling frequent reconfigurations to maintain connectivity in large systems without excessive time loss.
Solution Approach 2:
The metabase serves as a self-contained repository of pre-calculated paths that the system can autonomously query and select from during reconfiguration events. This self-service mechanism eliminates the need for external intervention or complex real-time computation, allowing rapid automated reconfiguration to maintain connectivity.
Data Source
AI summary
Systems and methods are provided for supporting efficient reconfiguration of an interconnection network having a pre-existing routing. An exemplary method can provide a plurality of switches, a plurality of end nodes, and one or more subnet managers, including a master subnet manager. The method can calculate, via the master subnet manager, a first set of one or more leaf-switch to leaf-switch multipaths. The method can store this first set of one or more leaf-switch to leaf-switch multipaths at a metabase. The method can detect a reconfiguration triggering event, and call a new routing for the interconnection network. Finally, the method can reconfigure the network according to the new routing for the interconnection network.