PaperlayerView exact revision on arXiv

Source-linked AI summary

Cluster Content Caching: An Energy-Efficient Approach to Improve Quality of Service in Cloud Radio Access Networks

Zhongyuan Zhao, Mugen Peng, Zhiguo Ding, Wenbo Wang, H. Vincent Poor

arXiv:1603.07052v1cs.ITcs.NI

TL;DR

C-RANs face heavy fronthaul and backhaul traffic that increases power consumption and worsens QoS for real-time services. The paper proposes cluster content caching, derives effective-capacity and energy-efficiency expressions, and optimizes resource allocation and RRH association; simulations report gains up to 0.57 Mbit/s/Hz and 0.004 Mbit/Joule, enlarged to 0.95 Mbit/s/Hz and 0.0055 Mbit/Joule with optimization algorithms.

  • Problem

    C-RAN content caching must reduce fronthaul and backhaul burdens while balancing centralized signal processing, distributed caching, interference-limited QoS analysis, and limited storage.

  • Method

    The paper proposes shared cluster caches for RRHs, derives tractable effective-capacity and energy-efficiency expressions, and jointly optimizes RRU allocation and RRH association.

  • Results

    0.57 Mbit/s/Hz and 0.004 Mbit/Joule are achieved with five required content objects, increasing to 0.95 Mbit/s/Hz and 0.0055 Mbit/Joule with the proposed optimization algorithms.

  • Takeaways & Limitations

    Cluster content caching improves QoS guarantees and energy efficiency by locally serving some requests while reducing backhaul loading.

Abstract

from arXiv · show

In cloud radio access networks (C-RANs), a substantial amount of data must be exchanged in both backhaul and fronthaul links, which causes high power consumption and poor quality of service (QoS) experience for real-time services. To solve this problem, a cluster content caching structure is proposed in this paper, which takes full advantage of distributed caching and centralized signal processing. In particular, redundant traffic on the backhaul can be reduced because the cluster content cache provides a part of required content objects for remote radio heads (RRHs) connected to a common edge cloud. Tractable expressions for both effective capacity and energy efficiency performance are derived, which show that the proposed structure can improve QoS guarantees with a lower power cost of local storage. Furthermore, to fully explore the potential of the proposed cluster content caching structure, the joint design of resource allocation and RRH association is optimized, and two distributed algorithms are accordingly proposed. Simulation results verify the accuracy of the analytical results and show the performance gains achieved by cluster content caching in C-RANs.

I. INTRODUCTION

The paper proposes cluster content caching in C-RANs to balance centralized signal processing with distributed caching while reducing backhaul loading and improving QoS. It analyzes performance and further optimizes RRU allocation and RRH association.

  • Motivation: C-RANs centralize BBUs for signal processing, but fronthaul and backhaul exchanges can burden real-time services and limit QoS guarantees.H-CRANs and EC-RANs are introduced as related architectures for balancing fronthaul and backhaul loading.
  • Motivation: Existing edge caching strategies are not directly applicable to C-RANs because cached content must pass through the cloud BBU pool before reaching RRHs.This creates additional content-delivery requirements between the cloud and RRHs.
  • Proposed structure: The proposed structure lets RRHs in the same cluster share a local cluster content cache, combining centralized signal processing with distributed caching.The paper analyzes the resulting backhaul-capacity constraint and associated delay improvement.
  • Analysis: Stochastic-geometry analysis derives tractable effective-capacity and energy-efficiency expressions for the cluster content caching structure.The analytical results quantify QoS and energy-efficiency gains when content requests can be served locally.
  • Results: 0.57 Mbit/s/Hz and 0.004 Mbit/Joule are the reported maximum improvements when five content objects are required.These values refer to effective capacity and energy efficiency, respectively, relative to the stated caching comparison.
  • Optimization: Joint RRU allocation and RRH association are optimized through a nested coalition formation game, with Shapley value used to reduce RRU-allocation complexity.The proposed optimization algorithms enlarge the reported gains to 0.95 Mbit/s/Hz and 0.0055 Mbit/Joule, respectively.

A. The Theory of Effective Capacity in Wireless Channels

Effective capacity is used to connect wireless-channel service capacity with QoS constraints such as delay and queue violations. The paper extends this framework to multi-hop C-RAN content delivery, showing how local caching changes backhaul loading and QoS performance.

  • Definition: Effective capacity characterizes the maximum constant arrival rate supported by a wireless channel under a specified QoS guarantee.The framework assumes an infinite queue and constant source arrival rate.
  • Definition: The QoS exponent θ controls tail decay: smaller θ represents a looser QoS requirement, while larger θ represents a stricter one.Queue and delay violation probabilities are related to θ through exponential tail behavior.
  • Derivation: Under block fading, effective capacity can be further derived because each channel coefficient remains constant within an RRU.This provides the channel-level expression used in the subsequent analysis.
  • C-RAN application: Content transmissions form multi-hop tandem links from backhaul or cluster cache to the edge cloud, then through fronthaul and radio access channels to users.Locally obtained content has effectively unbounded arrival rate, so its arrival-process delay can be ignored in the approximation.
  • Caching impact: Cluster content caching reduces backhaul loading and improves QoS guarantees by serving some objects locally.When cloud-caching QoS approaches cluster-caching QoS, the required backhaul capacity approaches infinity.
  • Caching impact: The structure can improve QoS with low power consumption even when requested content includes non-cacheable objects.Caching popular objects locally mitigates loading on links between local cloud centers and data centers, reducing delay for non-cacheable content.

III. PERFORMANCE ANALYSIS BASED ON STOCHASTIC GEOMETRY

The analysis models RRHs and users with homogeneous Poisson point processes and uses SINR quantization to derive tractable effective-capacity expressions under interference. The approximation is reported to match Monte Carlo results when configured with sufficiently fine intervals.

  • RRH and user locations are modeled as homogeneous marked Poisson point processes, with user marks representing requested content types.The model assumes independent requested content objects and uses PPPs to obtain tractable analytical results.
  • The analysis does not include coordinated multiple-point transmissions or network beamforming, while cache coordination uses a common cache in each cluster.This defines the scope of the stochastic-geometry model.
  • Effective capacity is derived by approximating the expectation of a SINR-dependent term through N quantization intervals between 0 and γmax.Each interval uses a representative SINR value, with midpoint quantization specified by the analysis.
  • Theorem 3 provides a tractable effective-capacity expression for a typical user accessing a specific RRH serving its requested content.The expression depends on the QoS exponent and the user’s distance to the serving RRH.
  • The SINR outage probability is derived from the PPP model and substituted into the effective-capacity expression.The derivation uses the probability generating functional of the PPP and independent exponential channel gains.
  • When γmax = 5×10^4 and N = 10^6, the analytical results match Monte Carlo results perfectly in the reported simulations.The paper attributes approximation accuracy to quantization theory and finer interval division.

B. Average Effective Capacity of A Typical Cluster CT

The cluster’s average effective capacity is obtained by weighting content-specific capacities by content popularity and distinguishing service through the local cluster cache from service through the cloud cache. The resulting analysis shows that effective-capacity gains increase with the cluster-cache hit ratio.

  • Users access the nearest RRH serving their desired content object so that received signal power is maximized.The derivation accounts for content-specific RRH partitions and nearest-serving-RRH distance distributions.
  • The average effective capacity of a typical cluster is formed by weighting each content object’s average capacity by its request popularity.The popularity ratios sum to one, and the hit ratio measures the probability that requests are satisfied by the local cache.
  • RRHs serving each content form independent thinned PPPs with densities λ_l whose sum equals the overall RRH density λ_R.This partition supports tractable content-specific capacity expressions.
  • Corollary 4 gives the average effective capacity of a typical cluster when users associate with the nearest RRH serving their desired content.The result follows from the content-popularity weighting and the stochastic-geometry model.
  • Average capacity is separated into local-cache and cloud-cache service components, with distinct QoS exponents for the two paths.The content-specific capacities are combined according to popularity and cache service location.
  • The effective-capacity gain ΔE_T increases as the local cluster-cache hit ratio P_hit increases.When all required content can be served locally, the paper states that local caching can approach its performance limits.

C. Energy Efficiency Performance of A Typical Cluster CT

Energy efficiency is defined as average effective capacity divided by average total power consumption, including radio, processing, storage, and backhaul components. Because local storage consumes much less power than backhaul retrieval, the cluster-cache structure is presented as energy-efficient for improving QoS guarantees.

  • Energy efficiency η_T is defined as average cluster effective capacity divided by average total power consumption.The total includes radio transmission, baseband processing, local cache storage, and backhaul acquisition power.
  • The power comparison includes the radio-processing cost per RRH, cluster-cache storage cost, cache size, and backhaul power consumption.These terms determine the total power used by the cluster content caching structure.
  • P_BH is larger than 10 W while P_CC is smaller than 1 W, supporting lower local-storage power cost than backhaul retrieval.The paper combines this power disparity with higher effective capacity to characterize cluster caching as energy efficient.

IV. JOINT DESIGN OF RRU ALLOCATION AND RRH ASSOCIATION BASED ON A NESTED COALITION FORMATION GAME

The joint management of RRU allocation and serving-RRH association is formulated as a coupled integer program and addressed through nested coalition-formation games. RRHs are grouped according to content service, with payoffs tied to effective-capacity improvement.

  • IV. JOINT DESIGN OF RRU ALLOCATION AND RRH ASSOCIATION BASED ON A NESTED COALITION FORMATION GAME: The joint RRU-allocation and RRH-association problem is NP-hard because the two radio-resource decisions are coupled.Both decisions are reformulated as coalition-formation games to obtain efficient solutions.
  • IV. JOINT DESIGN OF RRU ALLOCATION AND RRH ASSOCIATION BASED ON A NESTED COALITION FORMATION GAME: The association and allocation framework targets QoS guarantees by jointly managing scarce RRHs and radio resources.The paper motivates the joint formulation through the resource competition among users.
  • A. Serving RRH Association Strategy: RRHs in a cluster are partitioned into disjoint sets, each transmitting a corresponding content object from the requested content set.The partition is denoted by Π = {R_j1, ..., R_jM}.
  • A. Serving RRH Association Strategy: Users requesting the same content are collected into a corresponding set W_jm served by RRHs in R_jm.This links content demand to the RRH coalition responsible for transmission.
  • A. Serving RRH Association Strategy: The effective capacity of a content object served by its RRH set depends on the locations of serving RRHs and requesting users.The formulation uses these locations together with the serving-user distance.
  • A. Serving RRH Association Strategy: The QoS guarantees of requesting users are mainly determined by their serving RRHs and the resulting path loss characterized by distance.The model distinguishes content served by the local cluster cache from content served by the cloud cache through their QoS exponents.
  • A. Serving RRH Association Strategy: Serving-RRH association is modeled as a coalition-formation game in which RRHs are players and each partition represents an association result.The coalition payoff is defined by the effective-capacity improvement contributed by an RRH.

1) Utility Function Formulation:

Serving RRH association is modeled as a hedonic coalition formation game, while RRU allocation assigns content objects sharing a common RRU. The association algorithm negotiates coalition moves until reaching a Nash-stable partition.

  • 1) Utility Function Formulation:: The payoff of an RRH joining a coalition depends only on that coalition’s members, enabling a hedonic preference relation.An RRH compares alternative coalitions using the resulting individual and coalition utilities.
  • 2) A Distributed Coalition Formation Algorithm:: The association procedure initializes disjoint RRH coalitions and lets RRHs negotiate moves to other coalitions when preference criteria are satisfied.The process terminates when coalition membership no longer changes.
  • 2) A Distributed Coalition Formation Algorithm:: Algorithm 1 converges from any initial partition to a Nash-stable partition.This convergence result is stated as Proposition 5.
  • B. RRU Allocation Strategy: Content objects are treated as players in RRU allocation, forming disjoint coalitions whose members share a common RRU.The payoff of an RRU coalition is defined through the cluster’s sum effective capacity.

1) Utility Function Formulation:

The coupled RRU-allocation and RRH-association design uses coalition utilities, merge-and-split operations, and a nested algorithm. Fixed initialization and negotiation order support convergence to a Dhp-stable partition while unnecessary RRHs enter sleep mode.

  • 1) Utility Function Formulation:: The RRU-allocation utility includes expected effective capacity and a coalition-formation cost controlled by cRU.The cost term uses the coefficient cRU to regulate its impact on utility.
  • 1) Utility Function Formulation:: RRU allocation is non-superadditive with an empty core, so the grand coalition cannot be formed.The payoff depends on partition-wide spectral efficiency and includes coalition costs.
  • 2) A Distributed Merge and Split Algorithm:: Content-object coalitions are formed using merge and split operations that compare coalition utilities.Merge combines coalitions when joint utility is higher, while split separates a coalition when a partition improves utility.
  • C. A Nested Coalition Formation Game-Based Algorithm: The nested algorithm couples RRU allocation with RRH association by embedding the association procedure within each content-coalition iteration.Equation (39) obtains RRU-allocation utility from RRH-association utility, with fixed initialization and negotiation order.
  • C. A Nested Coalition Formation Game-Based Algorithm: The algorithm initializes disjoint content coalitions, applies merge and split operations, and terminates when coalition memberships stop changing.Each operation evaluates utilities using the nested association procedure.
  • C. A Nested Coalition Formation Game-Based Algorithm: After convergence, RRHs not required as nearest serving RRHs for users are placed into sleep mode to improve energy efficiency.This is performed after the final partition result is obtained.
  • C. A Nested Coalition Formation Game-Based Algorithm: The joint algorithm converges to a Dhp-stable partition, meaning no possible partition recurs through additional merge or split operations.The proof relies on increasing total utility and fixed initialization and negotiation order.
  • C. A Nested Coalition Formation Game-Based Algorithm: Distributed allocation requires only local information, while dynamic clustering, inter-cluster coordination, and content correlation remain potential global extensions.The paper identifies these extensions as ways to further improve cluster content caching.

V. A SUBOPTIMAL ALGORITHM WITH LOWER COMPUTATIONAL COMPLEXITY

The suboptimal algorithm reduces computational complexity by decoupling RRU allocation from RRH association and using an allocation utility independent of the serving-RRH partition.

  • V. A SUBOPTIMAL ALGORITHM WITH LOWER COMPUTATIONAL COMPLEXITY: The suboptimal RRU-allocation algorithm decouples RRU allocation and RRH association to reduce computational complexity.Its utility does not depend on the partition result of serving RRH association.
  • V. A SUBOPTIMAL ALGORITHM WITH LOWER COMPUTATIONAL COMPLEXITY: The decoupled formulation replaces the nested problem with two independent coalition formation games.This avoids repeatedly solving the RRH-association game within RRU-allocation iterations.

A. Utility Function Formulation Based on Shapley Value

The Shapley-value formulation evaluates each RRH’s contribution to transmitting a content object and uses differences in these values to manage sharing conflicts during RRU allocation.

  • A. Utility Function Formulation Based on Shapley Value: The Shapley value measures an RRH’s expected marginal contribution to transmitting a specific content object.The expectation is taken over the order in which RRHs join the grand coalition.
  • A. Utility Function Formulation Based on Shapley Value: A higher Shapley value indicates that an RRH is more important for the corresponding content object.These values quantify RRH importance separately for each content object.
  • A. Utility Function Formulation Based on Shapley Value: RRU-sharing utility increases as Shapley-value differences between content objects reduce their RRH-association conflicts.Identical Shapley values represent the most competitive case and yield zero payoff for sharing.

B. A Suboptimal RRU Allocation Algorithm

The suboptimal allocation approach decouples joint RRH association and RRU allocation into two coalition formation games, reducing complexity while preserving a stable partition for RRU allocation.

  • B. A Suboptimal RRU Allocation Algorithm: RRH association follows Algorithm 1, after which RRHs not required by any user are placed into sleep mode.The procedure first forms coalitions, then associates RRHs and turns off unnecessary RRHs.
  • B. A Suboptimal RRU Allocation Algorithm: Algorithm 3 decouples joint RRH and RRU allocations into two independent coalition formation games.This provides a suboptimal solution with lower computational complexity.
  • B. A Suboptimal RRU Allocation Algorithm: The Shapley value approximates RRU allocation with computational complexity O(D^3) in the studied problem.The passage notes that Shapley-value computation is NP-hard but can be approximated efficiently and accurately.
  • B. A Suboptimal RRU Allocation Algorithm: The RRU allocation game partitions N RRUs among L content objects with estimated complexity O((1 + b)L^2(2+b)L), where N = bL and 0 < b < 1.This complexity estimate applies to the hedonic coalition formation formulation for RRU allocation.
  • B. A Suboptimal RRU Allocation Algorithm: The joint design uses effective capacity-derived utility functions, so allocation reflects both channel capacity and QoS.The resource-allocation results are jointly determined by channel capacity and the QoS metric.

VI. SIMULATION RESULTS

The simulations validate the analytical results and show that cluster content caching improves effective capacity and energy efficiency, with further gains from optimized RRU allocation and RRH association.

  • Analytical effective-capacity results match Monte Carlo simulations perfectly, validating the theoretical derivations.
  • 0.57 Mbit/s/Hz and 0.004 Mbit/Joule are achieved over no caching when K = 5.These gains are attributed to more requests being served locally with shorter delay.
  • 0.95 Mbit/s/Hz and 0.0055 Mbit/Joule are reached with the proposed RRU allocation and RRH association optimization when K = 5.
  • Algorithm 2 achieves the best average effective capacity among the compared RRU allocation and RRH association schemes.The comparison includes orthogonal RRU allocation, RRU full-reusing, and another proposed algorithm.
  • The effective-capacity-to-power-consumption ratio rises at low RRH density but decreases at high RRH density.
Loading 1603.07052v1…