PaperlayerView exact revision on arXiv

Source-linked AI summary

A Traffic Load Balancing Framework for Software-defined Radio Access Networks Powered by Hybrid Energy Sources

Tao Han, Nirwan Ansari

arXiv:1407.7780v3cs.NI

TL;DR

Dense HetNets need traffic balancing that accounts for green-energy availability as well as network performance. The paper proposes a SoftRAN-based virtualized association scheme, reporting 30% lower on-grid power than LA at an 8% increase in average traffic delivery latency.

  • Problem

    Existing traffic load-balancing approaches do not treat green-energy utilization as a performance metric, despite its potential to reduce on-grid power use.

    Source: §II · p. 2

  • Method

    vGALA uses the SoftRAN controller to virtualize distributed user-association adjustments and balance latency against green-energy use.

    Source: §I · p. 2, §IV · p. 5

  • Results

    30% less on-grid power than LA came with an 8% increase in average traffic delivery latency under vGALA.

    Source: A. Performance comparison · p. 11

  • Takeaways & Limitations

    The trade-off between network performance and on-grid power consumption is adjustable per BS and controllable by the RAN controller.

    Source: §VII · p. 13

    The latency indicator is derived under an M/G/1 processor-sharing model, although practical traffic arrivals may be non-Poisson.

    Source: D. The generalization of vGALA · p. 8

Abstract

from arXiv · show

Dramatic mobile data traffic growth has spurred a dense deployment of small cell base stations (SCBSs). Small cells enhance the spectrum efficiency and thus enlarge the capacity of mobile networks. Although SCBSs consume much less power than macro BSs (MBSs) do, the overall power consumption of a large number of SCBSs is phenomenal. As the energy harvesting technology advances, base stations (BSs) can be powered by green energy to alleviate the on-grid power consumption. For mobile networks with high BS density, traffic load balancing is critical in order to exploit the capacity of SCBSs. To fully utilize harvested energy, it is desirable to incorporate the green energy utilization as a performance metric in traffic load balancing strategies. In this paper, we have proposed a traffic load balancing framework that strives a balance between network utilities, e.g., the average traffic delivery latency, and the green energy utilization. Various properties of the proposed framework have been derived. Leveraging the software-defined radio access network architecture, the proposed scheme is implemented as a virtually distributed algorithm, which significantly reduces the communication overheads between users and BSs. The simulation results show that the proposed traffic load balancing framework enables an adjustable trade-off between the on-grid power consumption and the average traffic delivery latency, and saves a considerable amount of on-grid power, e.g., 30%, at a cost of only a small increase, e.g., 8%, of the average traffic delivery latency.

I. INTRODUCTION

Dense HetNets need user association that balances traffic across heterogeneous base stations, but existing load-balancing approaches do not treat green-energy use as a performance metric. The paper frames this problem in a SoftRAN-managed network and models downlink traffic at its base stations.

I. INTRODUCTION

  • Dense SCBS deployments improve spectrum efficiency but magnify aggregate power use, while heterogeneous transmit powers can make SINR/RSSI association leave traffic imbalanced; SoftRAN offers centralized coordination.The RANC has a global view of network resources and traffic loads.Source: §I · p. 1

II. RELATED WORKS

  • Existing load-balancing approaches optimize network utilities without treating green-energy use as a performance metric; prior renewable-aware work includes BS power control and joint latency/green-energy association.Source: §II · pp. 2–3

III. SYSTEM MODEL AND PROBLEM FORMULATION

  • The modeled HetNet contains MBSs and SCBSs powered by both grid and green energy, with the RANC using global traffic-load and energy information to optimize association.Source: §III · p. 3, Figure 1 · p. 3

A. Traffic model

  • Traffic arrivals are modeled as Poisson with generally distributed packet sizes; each BS is modeled as an M/G/1 processor-sharing queue, and L(ρ_j) represents average traffic delivery latency.Source: A. Traffic model · pp. 3–4

B. Energy model

The energy model combines solar-derived green energy with on-grid backup and simplifies energy availability over each traffic-balancing interval. It represents BS power as a static component plus load-dependent dynamic consumption.

B. Energy model

  • Each MBS and SCBS has a solar panel; a charge controller schedules green-energy use based on solar intensity, BS consumption, and grid energy prices.The controller may use forecasts over a longer horizon to set green-energy availability for a balancing procedure.Source: B. Energy model · p. 4
  • Available green energy is held constant during a traffic-balancing procedure, and residual green energy is not redistributed in the model.The stated rationale is that balancing operates over minutes while solar generation is modeled over hours.Source: B. Energy model · p. 4
  • BS power is modeled as static consumption without traffic plus dynamic consumption caused by traffic, approximated as linear in load.The model can represent MBSs or SCBSs by adjusting their static-power and load-power parameters.Source: B. Energy model · p. 4

C. Problem formulation

The association formulation balances latency and on-grid power by weighting BS latency indicators according to green-energy availability. It also sets per-BS energy-latency preferences and restricts association using uplink pathloss.

C. Problem formulation

  • The objective balances the sum of BS latency indicators against on-grid power use, with κ adjusting the latency weight to control the trade-off.Source: C. Problem formulation · pp. 4–5
  • Green traffic capacity is the maximum load supportable by green energy; the energy-latency coefficient θ_j shifts desired load from actual load toward that capacity.θ_j = 0 is latency-sensitive, while θ_j = 1 makes desired load dominated by green traffic capacity.Source: C. Problem formulation · p. 5
  • A user may associate only with BSs whose uplink pathloss is below a predefined threshold.Source: C. Problem formulation · p. 5
  • BS latency weights reflect green-energy sufficiency: a weight above 1 gives that BS higher priority in reducing its latency indicator.Source: C. Problem formulation · p. 5

IV. VGALA: A GREEN ENERGY AND LATENCY AWARE LOAD BALANCING SCHEME

vGALA uses the RANC to optimize user association through iterative updates between virtual users and virtual BSs. Its user-side and BS-side algorithms run within the controller, locally updating information rather than exchanging it over the air interface.

IV. VGALA: A GREEN ENERGY AND LATENCY AWARE LOAD BALANCING SCHEME

  • The scheme has three phases: network measurement, association optimization and BS-status derivation, then association decisions based on optimized statuses and user data rates.The analysis assumes the RANC successfully collects network information and user data rates remain unchanged during an association process.Source: §IV · p. 5

A. The vGALA user association scheme

  • Virtual users and vBSs run the two algorithmic parts locally in the RANC, iterating user selections and BS operation-status updates until convergence.The virtualization covers computation resources, not all BS and user functions.Source: A. The vGALA user association scheme · pp. 5–6

1) The user side algorithm:

  • Each virtual user selects a vBS using operation status, perceived data rate and a pathloss-based eligibility set; the individual user-side computation is O(|B|).BSs outside the set whose uplink pathloss meets the user's threshold are effectively excluded from association.Source: 1) The user side algorithm · p. 6

2) The BS side algorithm:

  • Each vBS updates its traffic load as a weighted blend of its current load and perceived load, with δ(k) calculated by the RANC; the BS-side loop's runtime depends on convergence.The BS-side algorithm updates operation status from traffic load; its while-loop can take longer to find δ(k) when ψ(ρ(k)) is closer to the optimal value.Source: 2) The BS side algorithm · p. 6

B. The convergence of vGALA

The convergence analysis establishes that vGALA's traffic-load vector approaches a feasible limit and that the corresponding BS operation statuses converge. It also bounds the iterations for ε-suboptimality and identifies parameter choices that can speed convergence.

B. The convergence of vGALA

  • The original feasible set is nonconvex because associations are binary, so the proof first relaxes association variables to the interval [0,1].The relaxed feasible set is convex, and ψ(ρ) is strongly convex over it.Source: B. The convergence of vGALA · p. 7
  • The traffic-load vector converges to a vector ρ* in the feasible set, and the corresponding vBS operation statuses converge.The convergence argument uses the descent property of M(ρ(k)) when M(ρ(k)) differs from ρ(k).Source: B. The convergence of vGALA · p. 7
  • The iteration count for ε-suboptimality is bounded, with z = 1 − min{2qς, 2qςξ/Q} < 1 and ρ(1) as the initial traffic-load vector.The bound's stated target is ψ(ρ(k + 1)) − ψ(ρ*) ≤ ε.Source: B. The convergence of vGALA · p. 7
  • ψ(ρ) converges at least geometrically; for a fixed initial gap and ε, smaller z means faster convergence, and choosing ς and ξ appropriately can reduce iterations.Optimizing ς and ξ is outside the paper's scope.Source: B. The convergence of vGALA · p. 7

C. The optimality of vGALA

The paper establishes that, when the feasible set is nonempty and the traffic-load vector converges, the associated user association minimizes ψ(ρ).

C. The optimality of vGALA

  • If F is nonempty and the traffic-load vector converges to ρ*, the user association corresponding to ρ* minimizes ψ(ρ).The paper states that the converged vector's corresponding association minimizes the objective.Source: C. The optimality of vGALA · p. 7

D. The generalization of vGALA

The framework generalizes the network-performance term beyond the latency indicator. The paper shows vGALA can solve the generalized association problem under stated conditions on that metric.

D. The generalization of vGALA

  • The generalized model represents each BS's network performance by a function f(ρj) of its traffic load, rather than requiring the latency indicator.The paper notes that traffic arrivals may be non-Poisson and operators may choose other performance metrics.Source: D. The generalization of vGALA · p. 8
  • If f(ρj) is positive, convex and nondecreasing, the generalized objective is strongly convex over the relaxed feasible set.The paper derives convexity of the weighted objective under these conditions.Source: D. The generalization of vGALA · p. 8
  • Under the same positivity, convexity and monotonicity conditions on f(ρj), vGALA can solve the generalized user-association problem.Source: D. The generalization of vGALA · p. 8

V. THE PRACTICALITY OF THE VGALA SCHEME

vGALA uses location-based network inputs and RANC-controlled association updates, with adjustable energy–latency priorities. Admission control maintains feasibility under overload, preserving the scheme’s convergence and optimality properties.

A. Put into practice

  • The practical process collects location-based traffic density and green-energy information, optimizes BS operation status in the RANC, then determines associations centrally or through BS broadcasts.BSs can report updated location-based traffic density after user selections change.Source: A. Put into practice · pp. 8–9
  • Users lack BS traffic-load information, so choosing the BS with the highest data rate may lead to congestion rather than better QoS, limiting incentives to falsify reports.The scheme’s association rule uses both data rate and a factor reflecting BS load and available green energy.Source: A. Put into practice · p. 9

B. The energy-latency trade-off adaptation

  • θ lets each BS set its energy–latency sensitivity, while the RANC uses κ to set the adaptation range, from latency-aware behavior at κ = 0 toward energy-aware behavior as κ increases.The RANC may lower κ under heavy traffic to prioritize latency or raise it under light traffic to emphasize green-energy use.Source: B. The energy-latency trade-off… · p. 9

C. Admission control mechanism

  • Admission control assigns each location a BS-independent admission probability µ(x) to restrict loads when the UA problem would otherwise be infeasible; vGALA still converges to an optimal solution.The mechanism is intended to preserve feasibility under very heavy traffic, not to reduce energy consumption or latency.Source: C. Admission control mechanism · pp. 9–10

VI. SIMULATION RESULTS

In downlink HetNet simulations, vGALA balances on-grid power consumption against traffic delivery latency and adjusts that trade-off through κ and θ. Compared with the latency-aware scheme, it uses 30% less on-grid power for 8% higher average traffic delivery latency.

VI. SIMULATION RESULTS

  • The system-level downlink simulations randomly deploy three MBSs and seven SCBSs in a 2000m × 2000m area, with Poisson arrivals averaging 200 and 250-kbit traffic.The area is divided into 40,000 locations, each 10m × 10m.Source: §VI · p. 10

A. Performance comparison

B. Performance adaptation

  • With θ = 1, varying κ shifts vGALA from LA-like behavior as κ → 0 toward GA-like behavior as κ → ∞; κ sets the adaptation range, while θ adjusts performance within it.The optimal settings for θ and κ are beyond the paper’s scope.Source: Figure 6 · p. 11, B. Performance adaptation · p. 11

A. Performance comparison

  • For the fifth BS, GA directs more traffic to its green-energy supply, LA reduces its coverage area, and vGALA slightly reduces it to trade off power and latency.Larger coverage areas indicate that a BS serves more traffic.Source: A. Performance comparison · pp. 10–11
  • About 60 iterations are reported for convergence; the scheme virtualizes users and BSs in the RANC to avoid communication overhead over the air interface.The result is reported alongside the performance comparison.Source: A. Performance comparison · p. 11

C. Green energy generation rate evaluation

The evaluation shows how vGALA responds to green-energy availability and compares it with CRE approaches. The conclusion highlights its adjustable energy-latency trade-off and SoftRAN-based implementation.

C. Green energy generation rate evaluation

  • As solar-cell power efficiency rises, on-grid power consumption falls; latency behavior spans four regions, including local energy-latency trading at 16%–17%.With insufficient green energy, vGALA resembles LA; as availability grows, it redirects loads toward adequately powered BSs, and with sufficient energy at all BSs it again acts like LA.Source: C. Green energy generation rate… · pp. 11–12, Figure 8 · p. 12

D. Practicality evaluation

  • vGALA saves on-grid energy and reduces average traffic delivery latency versus CRE-LG, while increasing κ shifts its balance toward energy savings at a small latency cost.GA achieves the minimum on-grid energy consumption, but its average traffic delivery latency is significantly larger than other schemes.Source: Figure 9 · p. 13, D. Practicality evaluation · p. 13

VII. CONCLUSION

  • vGALA balances network performance against green-energy availability, with an adjustable trade-off controlled by the RANC and virtual users and BSs reducing over-the-air information exchanges.The paper also reports proving various properties, particularly convergence, and says simulations validate performance and practicality.Source: §VII · p. 13

APPENDIX A PROOF OF LEMMA 4

The proof establishes a termination condition for the BS-side algorithm and a geometric bound on objective-gap reduction. It then derives the iteration count needed for ε-optimality.

APPENDIX A PROOF OF LEMMA 4

  • The BS-side algorithm terminates with δ(k) = 0 or with 1 − δ(k) greater than ξ/Q.The proof establishes this termination condition when 0 ≤ 1 − δ(k) ≤ 1/Q.Source: Appendix A · p. 13
  • The objective gap contracts by a factor of at most 1 − min{2qς, 2qςξ/Q} per iteration.This bound is given for ψ(ρ(k + 1)) − ψ(ρ*) relative to ψ(ρ(k)) − ψ(ρ*).Source: Appendix A · p. 14
  • The proof derives the number of iterations required to reach ε-optimality by recursively applying the contraction bound.It sets z = 1 − min{2qς, 2qςξ/Q} and relates the remaining gap to ε.Source: Appendix A · p. 14
Loading 1407.7780v3…