TEC

TEC

Optimizing AI Service Placement and Resource Allocation in Mobile Edge Intelligence Systems — Lin et al. (2020) (arXiv:2011.05708, 2020)

What this evaluates

Evaluates the joint optimization of AI service placement and resource allocation in mobile edge computing by measuring the trade-off between computation time and energy consumption across varying network scales and task characteristics.

Datasets

• (no dataset; pure metric skill)

Metrics

• TEC (primary) — range: other
  • Weighted sum of total computation time and total energy consumption: $\beta \times \text{Time} + (1-\beta) \times \text{Energy}$, where $\beta$ is a user-defined weighting parameter balancing the two objectives.

Input / output format

Input: System configuration parameters including number of users $K$, program size $S$, per-user task data size $I_k$, channel gains $g_k$ and $h_k$, computing workload $C$, bandwidth $W$, noise power $N_0$, and power constraints.

Output: Per-user decisions on service placement (local compute vs. offload), uplink/downlink bandwidth allocation fractions, and edge CPU frequency allocation.

Scoring recipe

def compute_tec(total_time, total_energy, beta):
    tec = beta * total_time + (1 - beta) * total_energy
    return tec

Common pitfalls

• The TEC metric is a weighted sum, not a physical quantity; changing $\beta$ drastically shifts the optimal scheme and trade-off curve.
• The 'Optimal' baseline is only computationally tractable for small $K$ (≤10); comparisons for larger $K$ must exclude it due to exponential complexity.
• Channel models use specific Rayleigh fading parameters and a 0.75 correlation coefficient between uplink and downlink that must be replicated exactly for fair comparison.

Evidence (verbatim from paper)

In Fig. 4, we compare the TEC performance achieved by different schemes when the program size $S$ varies. Besides, we present the TEC performance comparison when the task data size $I$ varies in Fig. 5. From both figures, we observe that the TEC performance achieved by the proposed greedy search and ADMM-based methods are extremely close to the optimal scheme.

Citation

@misc{lin2020optimizing,
  title={Optimizing AI Service Placement and Resource Allocation in Mobile Edge Intelligence Systems},
  author={Lin et al. (2020)},
  year={2020},
  note={arXiv:2011.05708}
}

• arXiv: 2011.05708