Task Offloading and Scheduling in Fog RAN

Literature Review

Mobile computing has increasingly relied on offloading computational tasks from resource‑limited devices to proximate or remote computation infrastructures. Fog Radio Access Networks (fog RAN), which integrate fog/edge computing with traditional wireless access, have emerged as a prominent paradigm to support low‑latency, computation‑intensive mobile services. A large body of work explores offloading mechanisms, resource allocation, and network–compute interactions, but challenges remain when communication and computation can occur in parallel, as well as when scheduling affects overall execution latency.

1. Computation Offloading in Fog and Edge Architectures

Early research on MEC and fog computing generally focused on relieving mobile devices through offloading tasks to nearby nodes. Studies such as Peng et al. and Ku et al. characterized fog‑enabled RANs as hierarchical systems where access points (APs) and cloud centers jointly process tasks. These systems enable parallel computing at tiered nodes, thereby reducing computation delay compared to cloud‑only offloading. Subsequent work extended this idea by modeling multi-user and multi-access‑point scenarios, emphasizing the need for flexible offloading decisions tailored to network conditions and computation capacities.

Several works (e.g., Chen et al., Xing et al.) optimized offloading policies under constraints such as limited edge computation resources or user delay requirements. However, these analyses typically considered task placement alone and did not account for the impact of execution order at APs and cloud servers.

2. Parallel Communication and Computation in Fog RAN

A distinctive feature of fog RAN is that communication modules and computation units at APs operate independently, enabling tasks to be uploaded and executed in parallel. This is unlike earlier MEC architectures where communication bottlenecks often dictated execution timelines.

Research on parallelism in fog/edge environments (e.g., Lee et al., Liu et al.) employed queueing models or online optimization to examine interaction between communication pipelines and processing resources. Nevertheless, many assumed simple scheduling strategies such as first‑come, first‑served (FCFS), which can lead to suboptimal delay performance because parallel communication/computation pipelines require coordinated scheduling to avoid idle time or contention.

The interplay between uplink transmission, fronthaul delivery, and sequential CPU processing makes joint optimization significantly more complex.

3. Task Scheduling and Its Role in Delay Minimization

Beyond offloading decisions, task scheduling determines the order in which tasks are transmitted and processed at each computing node. In wireless MEC systems, scheduling affects:

• waiting time on uplink channels
• queue formation at edge servers
• contention on fronthaul links
• start times of cloud processing

Several works recognized that scheduling impacts delay. For example:

• Mao et al. explored joint scheduling and power allocation but did not incorporate parallel computing across nodes.
• Guo et al. and Ji et al. emphasized energy‑efficient resource allocation with sequential service models.

However, most studies implicitly assumed a single processing pipeline per node (i.e., no simultaneous communication and computation), limiting their applicability to fog RAN, where multiple operations may occur concurrently.

4. Joint Task Offloading and Scheduling

Because offloading and scheduling jointly determine execution delay, ignoring either can lead to suboptimal strategies. The problem is inherently a mixed integer optimization combining:

• binary offloading decisions (AP vs. cloud)
• real‑valued scheduling variables for communication and computation
• FCFS or other policies enforced at different stages

This formulation resembles a hybrid flow shop scheduling problem, which is computationally difficult and not widely addressed in communication systems. The complexity grows exponentially with the number of tasks.

Recent contributions, such as the work presented in the fog RAN study, highlight that:

• Offloading to different nodes enables parallel computing, improving throughput.
• Parallelism is amplified when communication and computation can overlap.
• Therefore, scheduling and offloading must be jointly optimized rather than separately determined.

The recursive computation offloading (RCO) approach aligns with this trend by incrementally building optimal solutions over task subsets while keeping complexity tractable.

5. Low-Complexity and Recursive Optimization Approaches

Exact solutions via branch‑and‑bound are feasible only for small task sets due to combinatorial blow‑up. To address this, researchers have proposed:

• heuristic scheduling
• greedy or priority-based selection
• semidefinite programming relaxations
• recursive decomposition techniques

The RCO algorithm represents a structured recursive strategy that leverages problem structure—namely, the ability to treat the “last scheduled task” separately and reuse subproblem results. This aligns with a broader movement in fog/MEC research to design scalable, near-optimal scheduling and offloading algorithms capable of exploiting multi-node parallelism.

6. Summary of Research Gaps and Motivations

Across the literature, several key limitations motivate unified frameworks such as those in fog RAN:

• Offloading alone is insufficient without coordinated scheduling.
• Parallelism across communication and computation is underutilized in most existing approaches.
• FCFS scheduling remains common but performs poorly when tasks have heterogeneous sizes and delays.
• Flow shop–style hybrid scheduling with parallel stages is rarely solved efficiently.

Consequently, the integration of task offloading and task scheduling—especially under parallel communication-computation architectures—fills an important gap in achieving delay‑effective computation offloading.