Distributed Processing Ring Sets for Network Contention Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed processing systems face inefficiencies due to poor network bandwidth utilization and network contention during collective operations, leading to reduced performance and limited functionality.
Innovation Solution
The implementation of a distributed processing system that utilizes ring sets with separate routers for each compute node, allowing for simultaneous data distribution without contention, and adapts algorithms like recursive doubling and hierarchical ring algorithms to optimize network utilization in a dragonfly network topology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional network topology is used for distributed processing, then system simplicity is maintained, but network bandwidth utilization is poor and network contention occurs during collective operations
Solution Approach 1:
The network is segmented into multiple dragonfly tiers with separate routers for each compute node, dividing the monolithic network into hierarchical segments that can operate independently, thereby reducing contention and improving bandwidth utilization during collective operations
Solution Approach 2:
The patent introduces a multi-dimensional dragonfly network topology with vertical and horizontal communication paths across multiple tiers, adding spatial dimensions to data flow that enable parallel communication channels and eliminate single-point bottlenecks
2Productivity
If compute nodes share common routers, then device complexity is reduced, but network contention increases during simultaneous data distribution
Solution Approach 1:
Each compute node is assigned a dedicated router, segmenting the network infrastructure so that data distribution operations from different compute nodes occur on separate physical paths, eliminating contention at shared router interfaces
Solution Approach 2:
The dedicated router acts as an intermediary between each compute node and the dragonfly network core, providing isolated access points that prevent direct contention while maintaining efficient routing through the multi-tier architecture
Data Source
AI summary
Various embodiments are generally directed to techniques for collective operations among compute nodes in a distributed processing set, such as by utilizing ring sets and local sets of the distributed processing set. In some embodiments, a ring set may include a subset of the distributed processing set in which each compute node is connected to a network with a separate router. In various embodiments, a local set may include a subset of the distributed processing set in which each compute node is connected to a network with a common router. In one or more embodiments, each compute node in a distributed processing set may belong to one ring set and one local set.