EdgeCampus wants to choose a workflow scheduler for faculty workflows running on its heterogeneous platform: 2 edge GPUs, 4 edge CPUs and up to 4 cloud VMs (rented per minute). Candidates: HEFT, PEFT and a budget-aware variant of HEFT (Chapter 6, short question).
Write an experiment plan (about one page):
- Research questions — at least two, precise and answerable.
- Workloads — which DAGs (random and real), with which parameter values (number of tasks, CCR, heterogeneity, budget levels)? How many instances?
- Platform model — how you model GPU vs CPU speed, edge–cloud communication, and cloud prices.
- Metrics — at least four, with definitions.
- Procedure and fairness — how you ensure a fair and reproducible comparison (seeds, validation, repetitions).
- Expected results and threats to validity — what you expect and why, and three things that could make your conclusions wrong.
Finally, here are (hypothetical) average results for 1,000 random DAGs with 100 tasks at CCR = 1. Interpret them in one paragraph:
| Algorithm | Avg SLR | Better than HEFT | Avg cost (¥) | Budget met |
|---|---|---|---|---|
| HEFT | 1.82 | — | 41.0 | 58% |
| PEFT | 1.74 | 64% | 43.5 | 52% |
| Budget-HEFT | 1.95 | 21% | 29.8 | 97% |