TL;DR
The paper introduces PSACCF, a novel algorithm for 5G network slice admission control that simultaneously balances service priority and resource fairness. By redefining fairness as the uniformity of gaps between service acceptance ratios, the authors achieve a 33.6% improvement in fairness and higher resource utilization compared to existing state-of-the-art methods.
What Problem Does This Paper Address?
In 5G and Beyond 5G (B5G) networks, "network slicing" allows a single physical infrastructure to support wildly different services—from high-bandwidth video streaming (eMBB) to ultra-reliable low-latency communications for autonomous vehicles (uRLLC).
The challenge lies in Slice Admission Control (SAC): when the network is congested, which service requests should be accepted? Traditionally, researchers have treated priority (favoring urgent services) and fairness (ensuring all services get a share) as separate or even contradictory goals. Furthermore, most existing models ignore "impatient" user behaviors—like balking (refusing to join a long queue) or reneging (leaving a queue after waiting too long)—which can drastically affect real-world network performance.
Methods
The authors propose the Prioritized Online Slice Admission Control Considering Fairness (PSACCF) algorithm. Because the underlying optimization problem is non-linear and non-convex, they developed a heuristic approach consisting of two main stages:
- Priority Amendment: The algorithm first ensures that high-priority services maintain a higher Cumulative Service Acceptance Ratio (CSAR) than lower-priority ones. It uses a metric called "Resource Efficiency" (the derivative of the acceptance ratio relative to resource consumption) to decide which requests to admit first to fix priority violations.
- Fairness Enhancement: Once priorities are satisfied, the algorithm optimizes fairness. Crucially, they redefine fairness using a modified Jain’s Fairness Index. Instead of aiming for equal acceptance across all services, they aim for uniform gaps between the acceptance ratios of different priority levels.
The model also explicitly incorporates mathematical functions to represent subscriber balking and reneging, making the simulation more representative of human-centric network usage.
Datasets
The paper does not use a public real-world dataset. Instead, the authors evaluated their algorithm using a custom-built C++ simulation environment.
- Service Types: They simulated four distinct service types (Service-1 to Service-4), ranging from elastic (flexible resource needs) to inelastic (strict requirements), and categorized them into "new requests" and "handoff requests" (which typically carry higher priority).
- Parameters: They used normalized resource vectors and arrival rates ($\lambda_0$) to stress-test the system under varying workloads.
Key Insights and Ideas
- Priority as a Gradient, Not a Wall: Rather than treating priority as an absolute "high-priority always wins" rule, the authors treat it as a relative difference in acceptance rates. This prevents low-priority services from being completely starved of resources.
- Fairness in Inequality: The "aha moment" here is the redefinition of fairness. In a prioritized system, true equality is impossible. By focusing on the uniformity of the gaps between priority levels, they create a system that feels "fairly unequal."
- Dynamic Resource Reservation: Unlike static systems, PSACCF's resource reservation adjusts automatically based on arrival rates. This flexibility allows the Slice Provider to maximize profit without violating Service Level Agreements (SLAs).
Key Results
- Fairness Boost: PSACCF demonstrated at least a 33.6% improvement in fairness degree compared to two common benchmarks: Maximum High Priority First (MHPF) and Aggressive High Priority First (AHPF).
- Resource Utilization: The algorithm achieved higher minimum average resource utilization (up to 13.4% better in specific scenarios) by better balancing the access opportunities across different service types.
- Stability: While aggressive preemption methods (like AHPF) showed "jittery" or unstable performance when users were prone to balking, PSACCF remained stable across different workloads.
- Efficiency: The computational overhead is minimal, with decision times fluctuating between 1ms and 4ms, making it suitable for real-time online deployment.
Limitations and Future Work
The authors acknowledge a few areas for further development:
- Parameter Sensitivity: The algorithm relies on two tuning parameters ($\epsilon$ and $\phi$) to balance priority and fairness. While this offers flexibility, finding the "sweet spot" for these parameters currently requires manual adjustment or simulation.
- Single Provider Focus: The current model assumes a single Slice Provider. The authors suggest that future work should explore multi-provider environments, where providers might compete for subscribers, leading to complex "gaming" behaviors that could impact admission logic.
- Profit vs. Fairness Trade-off: Pursuing extreme fairness can lead to a slight decrease in total profit (as higher-paying priority services are occasionally throttled to help lower-priority ones). Refining this economic balance is a suggested next step.
Summary updated