Summary
All sorts of useful tools for running and processing parameter sweep experiments with the dry core model in this repository.
Benchmarks
Some benchmarks for running the dry core model on various machines are provided below.
Cheyenne timing
| Truncation | Nodes | Days | Model time (with parallel) | Process time (with parallel) |
|---|---|---|---|---|
| T95 | 2 | 100 | 840s (1000s) | 150s (300s) |
| T95 | 1 | 100 | 1620s (1720s) | 300s (420s) |
Turned out that running process step in parallel was bad idea. Didn't really get much benefit.
Experiment storage space
| Truncation | Days | Space |
|---|---|---|
| T95 | 2000 | 120GB (1.2TB per series) |
| T95 | 5000 | 300GB (3TB per series) |
| T63 | 5000 | 130GB (1.3TB per series) |
Maximum timesteps with model stability
| Truncation | Timescale (days) | Maximum timesteps (s) |
|---|---|---|
| T95 | 0.2-4 | 400 |
| T95 | 10-40 | 600 (800?) |
| T95 | 100 | 800 |
| T95 | 200-400 | 1000 |
| T95 | 1000 | 1200 |
The 10-100 experiments were run with 600s timesteps. Haven't tested 800.