Cant-Be-Late Scheduling Problem
Cloud spot instance scheduling benchmark with multiple difficulty variants.
Variants
This problem has 12 variants combining three dimensions:
| Dimension | Options | Description |
|---|---|---|
| Availability | high_availability |
43-78% spot availability regions |
low_availability |
4-40% spot availability regions | |
mixed_availability |
Both high and low availability regions | |
| Deadline | tight_deadline |
52 hours (4-hour slack) |
loose_deadline |
70 hours (22-hour slack) | |
| Overhead | small_overhead |
0.05 hours (3 minutes) restart overhead |
large_overhead |
0.20 hours (12 minutes) restart overhead |
Variant Examples
| Variant | Deadline | Overhead | Regions |
|---|---|---|---|
high_availability_loose_deadline_large_overhead |
70h (22h slack) | 0.20h (12 min) | High (43-78%) |
high_availability_loose_deadline_small_overhead |
70h (22h slack) | 0.05h (3 min) | High (43-78%) |
high_availability_tight_deadline_large_overhead |
52h (4h slack) | 0.20h (12 min) | High (43-78%) |
high_availability_tight_deadline_small_overhead |
52h (4h slack) | 0.05h (3 min) | High (43-78%) |
low_availability_loose_deadline_large_overhead |
70h (22h slack) | 0.20h (12 min) | Low (4-40%) |
low_availability_loose_deadline_small_overhead |
70h (22h slack) | 0.05h (3 min) | Low (4-40%) |
low_availability_tight_deadline_large_overhead |
52h (4h slack) | 0.20h (12 min) | Low (4-40%) |
low_availability_tight_deadline_small_overhead |
52h (4h slack) | 0.05h (3 min) | Low (4-40%) |
mixed_availability_loose_deadline_large_overhead |
70h (22h slack) | 0.20h (12 min) | Mixed |
mixed_availability_loose_deadline_small_overhead |
70h (22h slack) | 0.05h (3 min) | Mixed |
mixed_availability_tight_deadline_large_overhead |
52h (4h slack) | 0.20h (12 min) | Mixed |
mixed_availability_tight_deadline_small_overhead |
52h (4h slack) | 0.05h (3 min) | Mixed |
Region Details
High Availability Regions (43-78% spot availability):
us-west-2b_k80_1(77.5%)us-west-2b_v100_8(66.3%)us-west-2a_v100_8(62.8%)us-west-2b_v100_1(43.2%)
Low Availability Regions (4-40% spot availability):
us-west-2a_k80_8(40.3%)us-west-2b_k80_8(39.1%)us-west-2a_v100_1(30.3%)us-west-2a_k80_1(4.3%)
Directory Structure
cant_be_late/
├── common/ # Shared evaluator code
│ ├── __init__.py # Region and config definitions
│ ├── run_evaluator.py # Main evaluation logic
│ ├── cbl_evaluator.py # Core simulation evaluator
│ ├── sim_worker.py # Simulation worker
│ ├── cant-be-late-simulator/ # Simulator code
│ └── real_traces.tar.gz # Trace data (extracted by set_up_env.sh)
├── {variant}/ # Each variant directory
│ ├── evaluator.py # Thin wrapper importing from common/
│ ├── readme # Problem specification and API
│ ├── config.yaml # Variant configuration
│ └── resources/ # Resources for evaluation
└── README.md # This file
Common Parameters
All variants share:
- Task duration: 48 hours
- On-demand price: ~$3.06/hr
- Spot price: ~$0.97/hr
- Time limit: 300 seconds