Autonomous Task Offloading of Vehicular Edge Computing with Parallel Computation Queues

This work considers a parallel task execution strategy in vehicular edge computing (VEC) networks, where edge servers are deployed along the roadside to process offloaded computational tasks of vehicular users. To minimize the overall waiting delay among vehicular users, a novel task offloading solution is implemented based on the network cooperation balancing resource under-utilization and load congestion. Dual evaluation through theoretical and numerical ways shows that the developed solution achieves a globally optimal delay reduction performance compared to existing methods, which is also validated by the feasibility test over a real-map virtual environment. The in-depth analysis reveals that predicting the instantaneous processing power of edge servers facilitates the identification of overloaded servers, which is critical for determining network delay. By considering discrete variables of the queue, the proposed technique’s precise estimation can effectively address these combinatorial challenges to achieve optimal performance.

💡 Research Summary

**
The paper addresses the challenge of task offloading in vehicular edge computing (VEC) systems where roadside edge servers are equipped with multiple CPUs and finite‑capacity queues. Existing works either assume single‑CPU servers or relax the discrete nature of queue lengths, thereby missing the sharp “threshold” behavior that occurs when the number of offloaded tasks exceeds the number of available CPUs. To capture this phenomenon, the authors explicitly model the queue length (n_a) at each RSU‑a and the CPU count (k_a). When (n_a \le k_a) the waiting delay is zero; once (n_a > k_a) tasks are queued and experience a cumulative waiting time that grows non‑linearly. Equations (3) and (5) provide closed‑form expressions for the offloading delay, including transmission, waiting, and execution components, based on a simple round‑robin load‑sharing policy across CPUs.

The offloading decision is represented by binary variables (x_{ia}) indicating whether vehicle‑i associates with RSU‑a (or processes locally when (a=0)). The global objective is to minimize the sum of all vehicle delays (\sum_i T_{ia}(x_{ia})). This yields a combinatorial optimization problem that is NP‑complete due to the multi‑CPU scheduling aspect. Rather than resorting to complex heuristics, the authors decompose the problem into sub‑problems that can be solved cooperatively by vehicles and RSUs through a message‑passing (MP) framework inspired by factor graphs and dynamic programming.

In the MP algorithm each vehicle and RSU maintains factor functions (Q_i(\cdot)) and (R_a(\cdot)). Messages (\alpha, \beta, \rho, \eta) convey “the (l)-th smallest value” of local cost tables, enabling each node to update its belief about the optimal association without central coordination. The authors prove two key theoretical results: (1) the MP updates converge in a finite number of iterations because the message space is finite and the updates are monotone; (2) the converged solution is globally optimal for the original discrete problem, leveraging the sub‑modular structure of the objective.

Extensive simulations explore a wide range of system parameters: number of CPUs per RSU, queue capacities, vehicle densities, and task sizes. Compared with baseline methods—continuous relaxations, deep‑reinforcement‑learning (DRL) based schemes, and simple heuristic allocations—the proposed approach reduces average waiting delay by 30 %–45 % and keeps queue overflow probability below 2 %. A digital‑twin testbed built on real‑map data validates the algorithm in a realistic mobility scenario, showing that the distributed MP protocol can adapt to vehicle handovers and RSU cooperation while maintaining low latency.

The paper also discusses limitations: the communication delay is modeled with a fixed data rate, ignoring fast fading and interference; the round‑robin CPU assignment, while simple, may not achieve perfect load balance under heterogeneous task profiles. Future work is suggested on adaptive scheduling that incorporates channel prediction, energy consumption constraints, and pipelined multi‑CPU execution for a single task.

In summary, the work makes three major contributions: (i) an exact discrete queue model for multi‑CPU edge servers that captures the abrupt increase in waiting time when servers become overloaded; (ii) a decentralized, message‑passing based offloading algorithm with provable convergence and global optimality; and (iii) thorough theoretical, simulation, and real‑world validation demonstrating superior delay performance over existing VEC offloading strategies. This positions the proposed technique as a practical solution for latency‑critical vehicular applications such as autonomous navigation and real‑time AI inference.