Collective Network Topology Selection for Low-Latency Offloading
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collective communications in large-scale networks face scalability issues due to increased communication costs and latency, which can be mitigated by offloading processing tasks to network adapters or switches managed by a collective engine.
Innovation Solution
A collective engine optimizes network topology and communication routes by evaluating multiple cost models to minimize latency and communication costs, using parameters such as number of ports, tree depth, and stride value to configure efficient communication paths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If the number of nodes in computing systems increases, then processing power and computing capability are improved, but scalability issues arise due to inefficiencies in data sharing and processing
Solution Approach 1:
The patent introduces a collective network as an intermediary layer between processing nodes. This collective network includes collective engines that manage and coordinate data sharing and processing operations across multiple nodes, enabling efficient communication and collaboration without requiring direct node-to-node interactions, thus resolving scalability limitations
2Productivity
If collective communications are performed across all processes/nodes, then global communication operations are achieved, but performance bottlenecks occur due to network communication costs
Solution Approach 1:
The patent segments the collective network into multiple collective engines, each responsible for managing specific groups of nodes or specific communication operations. This segmentation allows parallel processing of communication tasks, reducing the time required for global communication operations and eliminating performance bottlenecks
3Productivity
If communications processing is offloaded from nodes to network adapters/switches, then processor efficiency is improved, but device complexity increases
Solution Approach 1:
The collective engines implemented in network adapters or switches are designed to perform multiple functions including data routing, processing coordination, and communication management. This multi-functionality allows the network infrastructure to handle diverse communication patterns and operations without requiring separate dedicated hardware for each function, thereby limiting the increase in device complexity
Data Source
AI summary
The disclosed device includes a collective engine that can select a communication cost model from multiple communication cost models for a collective operation and configure a topology of a collective network for performing the collective operation using the selected communication cost model. Various other methods, systems, and computer-readable media are also disclosed.