RQ-SAFE Couples Request-Resource Scheduling to Cut Edge CPU Imbalance 6.1%

RQ-SAFE cuts CPU imbalance by 6.1% and peak CPU by 2.3% over a GNN-based baseline on public-mixed edge workloads, all without forcing a fixed service order through the network.

Most edge orchestration systems treat a service function chain (SFC) as an unchangeable sequence. RQ-SAFE's authors realized that intent-driven edge services let you locally reorder VNF segments inside a directed acyclic graph without changing what the service actually does. That freedom has been sitting on the table untouched.

The Orchestration Freedom Nobody Used

Existing methods optimize VNF placement, routing, or queue-aware scheduling but assume the service order is predetermined. RQ-SAFE couples request-side ordering with resource-side consequences. For each feasible local order, it previews what the current edge infrastructure will look like under that order, then uses the retained order to pick VNF instances and construct paths.

Queue state feeds into every stage: evaluating local orders, ranking per-VNF candidates, and doing final QoS validation. A profile-aware learning-assisted re-ranker then refines the top-K candidates to balance QoS objectives against resource load.

The Numbers: 6.1% Imbalance Reduction, 4.53 Point QoS Gain

On matched edge SFC-DAG workloads, RQ-SAFE matches graph-aware baselines on QoS compliance but wins on resource balance. The CPU imbalance drop of 6.1% and peak CPU reduction of 2.3% come with limited extra control-plane decision time.

Ablation results make the coupling concrete. Enabling both local-order flexibility and queue awareness improves QoS by 4.53 percentage points compared to disabling both. The interaction effect between the two factors adds 3.83 percentage points, meaning the combination is worth more than the sum of its parts.

RQ-SAFE runs against GNN-DAG-Score, a graph neural network baseline, and beats it on these metrics. That's not a casual benchmark - GNN-DAG-Score is itself a well-tuned approach for this problem class.

What This Changes

Intent-aware edge orchestrators now have a practical paradigm for exploiting request-side freedom that was previously academic. The next step is seeing how RQ-SAFE's preview-and-commit model holds up under rapidly shifting edge topologies and adversarial request patterns.

Source: RQ-SAFE: Coupled Request-Resource Scheduling for Online Edge SFC-DAGs
Domain: arxiv.org