Dimensionally-Adaptive Load-Balancing for Congested HyperX Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Adaptive routing algorithms in large-scale parallel computing systems often fail to guarantee in-order packet delivery, leading to additional reordering burdens on destination Processing Elements (PEs), and they do not effectively manage network congestion, which can hinder performance.
Innovation Solution
The implementation of a Dimensionally-Adaptive Load-Balancing (DAL) algorithm in high dimension symmetric HyperX networks, where routers use de-routed paths to avoid congestion while maintaining in-order packet delivery by dynamically selecting paths based on network load, ensuring that packets are routed along the shortest path when possible and using de-routed paths only when necessary to avoid congested links.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If adaptive routing algorithms are used to dynamically avoid congested links, then network performance is improved, but in-order packet delivery is not guaranteed
Solution Approach 1:
The patent implements dynamic routing decisions at each router hop, where the routing algorithm adapts to current network conditions by selecting among multiple possible paths. The routing state is dynamically updated based on congestion detection, allowing the system to maintain in-order delivery while adapting to changing network conditions in real-time
Solution Approach 2:
The patent segments the routing decision-making process into discrete dimensional steps, where packets progress through dimensions in a controlled sequence. This segmentation of the routing path into manageable dimensional segments allows adaptive decisions at each step while preserving overall in-order delivery guarantees
2Productivity
If de-routed paths are used to avoid congestion, then load balancing is improved, but path length increases
Solution Approach 1:
The patent applies de-routing partially, only when congestion is detected in the current dimensional path. Rather than always using de-routed paths, the system uses them selectively and only to the extent necessary to avoid congestion, thereby achieving load balancing while minimizing the increase in path length
Solution Approach 2:
The patent changes routing parameters dynamically by switching between shortest-path routing and de-routed paths based on network congestion conditions. When congestion is detected, the routing parameters are adjusted to select alternative paths; when no congestion exists, the system reverts to optimal shortest paths, thus balancing load while controlling path length
Data Source
AI summary
Illustrated is a computer system and method that includes a Processing Element (PE) to generate a data packet that is routed along a shortest path that includes a plurality of routers in a multiple dimension network. The system and method further include a router, of the plurality of routers, to de-route the data packet from the shortest path to an additional path, the de-route to occur where the shortest path is congested and the additional path links the router and an additional router in a dimension of the multiple dimension network.