Copper Documentation
Copper is a read-only cooperative caching layer for scalable metadata and data reuse on large HPC systems. This documentation set is organized as a practical manual: start with the overview, move into build and platform guidance, then use the profiling and runtime behavior sections to study launches, scaling, and path usage.
Overview
Platforms and Build
Running and Analysis
- Launch and Analysis Runbook
- Purpose
- Step 1: Choose a Log Level
- Step 2: Choose an Address-Book Source
- Step 3: Choose Profiling Options
- Step 4: Launch Copper
- Step 5: Run the Workload
- Step 6: Stop Copper
- Step 7: Locate the Job Output Directory
- Step 8: Run Aggregate Profiling
- Step 9: Inspect Cache Table Usage
- Step 10: Run Path Usage Analysis
- Sample Outputs from Aurora Run 8405873
- Practical Reading Order
- Related Pages
- Profiling Overview
- Summary
- Profiling Controls
- Validation Scope
- Profiling Use Cases
- Validated Profiling Outputs
- Representative Iter3 Cluster Totals
- Produced Files
- Final Full-Path Profiling Findings
- Cross-Rank Stability
- What the Profiles Say About the Workload
- Concrete Hot-Path Examples
- Metric Reading Guide
- Startup Timing Versus Profiling
- Related Pages
- Profiling Reference
- Environment Path Analysis
Runtime Behavior and Scaling
- Metadata ENOENT TTL Evaluation
- Registration and Scaling
- Problem Statement
- Runtime Design
- Readiness Semantics
- Address-Book Preparation
- Retained Timing Signals
- Observed Scaling Behavior
- Current Stable High-Scale Reference
- Iter1 Versus Iter2 at High Scale
- Step-by-Step Runtime Sequence
- What the Code Measurement Means
- Operational Interpretation
- Operational Use
- Runtime Tuning and Notes
- Future Work