Fat-Tree Network Reconfiguration via Subnet Manager
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing systems, network reconfiguration is often costly and inefficient due to configuration-oblivious routing, which results in substantial modifications to existing routes and increased reconfiguration time, especially in large systems like Exascale computing environments where frequent reconfigurations are needed to maintain connectivity and performance.
Innovation Solution
The implementation of a subnet manager that computes new routing for interconnection networks, such as InfiniBand networks, while considering pre-existing routing to minimize modifications and efficiently reconfigure the network, using techniques like minimal routing updates and metabase-driven routing to preserve existing paths and reduce computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If configuration-oblivious routing is used for network reconfiguration, then new routes can be computed quickly, but substantial modifications to existing routes are required, increasing reconfiguration overhead and time
Solution Approach 1:
The patent pre-computes and stores alternative routes and path information in routing tables and metabases before reconfiguration events occur. When a reconfiguration event happens, the subnet manager can quickly retrieve pre-computed alternative routes that maintain connectivity, avoiding the need to compute entirely new routes from scratch. This preliminary preparation significantly reduces reconfiguration time while minimizing modifications to existing routes.
Solution Approach 2:
The patent implements dynamic route selection that adapts based on current network state. The subnet manager evaluates existing routes and selectively modifies only those paths that are affected by reconfiguration events, rather than reconfiguring all routes. This dynamic approach allows the system to maintain optimal performance by preserving working routes and minimizing changes, thereby reducing reconfiguration overhead.
2Reliability
If complete reconfiguration is performed to ensure connectivity, then network reliability is maintained, but reconfiguration overhead and computational cost increase significantly
Solution Approach 1:
The patent performs partial reconfiguration by identifying and modifying only the specific routes that are affected by reconfiguration events, rather than reconfiguring the entire network. The subnet manager analyzes the impact of events (such as link failures or node additions) and selectively updates routing tables for affected source-destination pairs. This partial action approach maintains network reliability for all connections while significantly reducing reconfiguration complexity and overhead compared to complete reconfiguration.
Solution Approach 2:
The patent segments the reconfiguration process into independent path computations for different source-destination pairs. Each route is evaluated and modified independently based on its specific requirements and the nature of the reconfiguration event. This segmentation allows parallel processing of route updates and reduces the overall complexity by breaking down the monolithic reconfiguration task into manageable, independent segments.
3Productivity
If routing is optimized for performance, then network throughput is improved, but reconfiguration becomes more costly when network topology changes
Solution Approach 1:
The patent implements feedback mechanisms where the subnet manager continuously monitors network performance metrics and topology changes. When reconfiguration events occur, the system uses feedback from existing route performance data to quickly determine optimal alternative paths. The metabase stores performance information about different routes, allowing the subnet manager to select alternative paths that maintain or improve throughput while minimizing reconfiguration time. This feedback-driven approach ensures that performance optimization does not come at the cost of excessive reconfiguration time.
Data Source
AI summary
Systems and methods are provided for supporting efficient reconfiguration of an interconnection network having a pre-existing routing comprising. An exemplary method can provide, a plurality of switches, the plurality switches comprising at least one leaf switch, wherein each of the one or more switches comprise a plurality of ports, and a plurality of end nodes, wherein the plurality of end nodes are interconnected via the one or more switches. The method can detect, by a subnet manager, a reconfiguration triggering event. The method can compute, by the subnet manager, a new routing for the interconnection network, wherein the computing by the subnet manager of the new routing for the interconnection network takes into consideration the pre-existing routing and selects the new routing for the interconnection network that is closest to the pre-existing routing. The method can reconfigure the interconnection network according to the new routing.