hyperpod-performance-debugger

Diagnose Amazon SageMaker HyperPod performance issues via NCCL bandwidth and filesystem metrics.

15|9|Updated Aug 16, 2025
One-click install
npx skills add https://github.com/haozhx23/HyperPod-InstantStart --skill hyperpod-performance-debugger
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: hyperpod-performance-debugger
Source: https://github.com/haozhx23/HyperPod-InstantStart/tree/main/.claude/skills/hyperpod-performance-debugger
Command: npx skills add https://github.com/haozhx23/HyperPod-InstantStart --skill hyperpod-performance-debugger

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires aws, ssm, jq, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill helps diagnose performance issues on Amazon SageMaker HyperPod clusters, ensuring efficient NCCL bandwidth and optimal filesystem throughput.

Core Features & Use Cases

  • Performance Diagnostics: Surface host-side signals for uneven NCCL bandwidth and filesystem performance.
  • Remediation Routing: Guide users to appropriate sibling skills for further investigation and resolution.
  • Automated Snapshots: Gather performance data using scripts/perf-snapshot.sh for quick analysis.
  • Use Case: If a user notices that their training jobs are taking longer than expected on certain nodes, this skill can be used to identify if the issue is related to uneven NCCL bandwidth or poor filesystem performance.

Quick Start

Run the performance snapshot script: bash scripts/perf-snapshot.sh --cluster <CLUSTER_NAME_OR_ARN> --region <REGION>

Frequently Asked Questions about hyperpod-performance-debugger

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagnose uneven NCCL bandwidth on SageMaker HyperPod clusters?

Run a performance snapshot script against your SageMaker HyperPod cluster to surface host-side signals for uneven NCCL bandwidth and identify nodes causing communication bottlenecks during distributed training jobs.

Why are my SageMaker HyperPod training jobs taking longer than expected?

Training jobs may take longer due to uneven NCCL bandwidth or poor filesystem performance. Diagnose the issue by running a performance snapshot to analyze these host-side metrics and pinpoint the affected nodes.

How do I check filesystem performance issues on Amazon SageMaker HyperPod?

Check filesystem performance on Amazon SageMaker HyperPod by executing an automated performance snapshot script that gathers metrics to identify throughput issues slowing down your distributed training jobs.

Do I need AWS CLI and jq to debug SageMaker HyperPod performance?

Yes, you need AWS CLI, jq, and SSM to debug SageMaker HyperPod performance, as the automated snapshot script relies on these dependencies to access cluster metrics and parse performance data.

What is the best way to troubleshoot NCCL bandwidth issues on AWS SageMaker?

The best way to troubleshoot NCCL bandwidth issues on AWS SageMaker is to run a performance diagnostics snapshot that analyzes host-side signals and routes you to appropriate remediation skills for resolution.