AI/ML Workload Graph Expansion for Collective Communication Testing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing graph-based representations of AI/ML workload execution in distributed computing scenarios lack detailed low-level processing instructions for collective communication operations, limiting the ability to test and optimize these operations effectively.

Innovation Solution

A method and system that expand collective communication nodes in a graph-based representation to include low-level processing instructions, such as send and receive primitives, allowing for detailed emulation and optimization of communication algorithms.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If collective communication operations are represented at high-level only, then the graph-based representation remains simple and manageable, but the ability to test and optimize low-level processing instructions is limited

Engineering Contradiction:

Improvegraph-based representation structureVSAvoidtesting capability for low-level processing

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The collective communication node is segmented into multiple individual communication operations at the low-level. Each high-level collective communication operation is broken down into fundamental send and receive primitives that can be independently analyzed and optimized, enabling detailed testing while maintaining overall system manageability

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system dynamically expands collective communication nodes based on testing requirements. When low-level optimization testing is needed, the nodes are expanded into detailed processing instructions; when high-level overview suffices, they remain aggregated, providing adaptability across different testing scenarios

Inventive Principle:

Principle #15Dynamics

2Adaptability or versatility

If collective communication nodes are expanded to low-level processing instructions, then detailed testing and optimization capability is enabled, but the complexity of the graph-based representation increases

Engineering Contradiction:

Improvetesting capability for low-level processingVSAvoidgraph-based representation structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The expansion of collective communication nodes is dynamic rather than static. The system can expand nodes to low-level processing instructions when detailed testing is required, and collapse them back to high-level representations when overview is sufficient, managing complexity adaptively based on testing needs

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The detailed low-level processing instructions are extracted from the high-level collective communication representation only when needed for specific testing scenarios. This allows the system to maintain a simple high-level graph for general purposes while accessing detailed low-level information on-demand for optimization testing

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20250328453A1Methods, systems, and computer readable media for emulating a distributed computing scenario using a graph-based representation of artificial intelligence/machine learning workload execution with an expanded collective communication operation

Publication Date: 2025.10.23 KEYSIGHT TECHNOLOGIES INC

AI summary

A method for emulating a distributed computing scenario using a graph-based representation of AI/ML workload execution with an expanded collective communication operation includes receiving a graph-based representation of AI/ML workload execution comprising a collective communication node and expanding the collective communication node by replacing a collective communication operation of the collective communication node with low-level processing instructions. A modified graph-based representation of AI/ML workload execution comprising the low-level processing instructions is generated. The modified graph-based representation of AI/ML workload execution is implemented in an emulated test case using an emulation engine.