PaperlayerView exact revision on arXiv

Source-linked AI summary

Toward Cloud-based Vehicular Networks with Efficient Resource Management

Rong Yu, Yan Zhang, Stein Gjessing, Wenlong Xia, Kun Yang

arXiv:1308.6208v1cs.DCcs.NI

TL;DR

The paper addresses how connected vehicular networks can handle traffic-related data and limited vehicle resources. It proposes a three-layer cloud architecture with game-theoretical allocation and reservation-based VM migration, reporting fast allocation convergence and reduced migrated-VM dropping rates. The design supports shared computation, storage, and bandwidth resources across vehicular, roadside, and central clouds.

  • Problem

    Connected vehicular networks must collect and store large amounts of traffic-related data, while individual vehicles have limited computation and storage resources for emerging applications.

  • Method

    The paper integrates vehicular, roadside, and central clouds and uses game theory for resource allocation plus resource reservation for mobility-related VM migration.

  • Results

    Resource allocation converges to a Nash equilibrium after nearly 10 iteration rounds, while resource reservation significantly reduces migrated-VM dropping rates.

  • Takeaways & Limitations

    The proposed architecture facilitates shared computation, storage, and bandwidth resources, with resource management addressing VM competition and mobility-related migration.

Abstract

from arXiv · show

In the era of Internet of Things, all components in intelligent transportation systems will be connected to improve transport safety, relieve traffic congestion, reduce air pollution and enhance the comfort of driving. The vision of all vehicles connected poses a significant challenge to the collection and storage of large amounts of traffic-related data. In this article, we propose to integrate cloud computing into vehicular networks such that the vehicles can share computation resources, storage resources and bandwidth resources. The proposed architecture includes a vehicular cloud, a roadside cloud, and a central cloud. Then, we study cloud resource allocation and virtual machine migration for effective resource management in this cloud-based vehicular network. A game-theoretical approach is presented to optimally allocate cloud resources. Virtual machine migration due to vehicle mobility is solved based on a resource reservation scheme.

I. INTRODUCTION

The paper motivates cloud-based vehicular networks by the resource limits of individual vehicles and proposes a hierarchical architecture that aggregates resources across vehicles, roadside infrastructure, and data centers. It then targets efficient resource allocation, service continuity, and reliability under vehicle mobility.

  • Connected vehicular networks can improve transport safety, congestion, pollution, and driving comfort through traffic-data gathering, processing, and dissemination.The passage gives highway safety warnings and real-time traffic information as examples.
  • Individual vehicles have limited computation, storage, and bandwidth resources, while emerging applications require complex computation and large storage.The paper identifies multimedia entertainment, vehicular social networking, and location-based services as examples.
  • The proposed hierarchical architecture aggregates redundant resources from data centers, roadside units, and vehicles into a pervasive cloud environment for mobile vehicles.The architecture organizes these resources into three layers so vehicles can select cloud services resiliently.
  • Central clouds provide abundant resources but higher communication delay, whereas roadside and vehicular clouds provide better communication quality with more limited resources.
  • The paper studies cloud resource allocation using a game-theoretical approach and addresses mobility-related virtual machine migration with resource reservation.The article structure assigns resource allocation to Section IV and mobility countermeasures to the proposed management strategies.

II. PROPOSED CLOUD-BASED VEHICULAR NETWORKS ARCHITECTURE

The proposed architecture is hierarchical, combining vehicular, roadside, and central clouds that vehicles access according to mobility and resource needs. It includes alternative vehicular-cloud customization strategies and trades global resource utilization against system overhead.

  • The architecture consists of three interacting layers: vehicular cloud, roadside cloud, and central cloud, which vehicles exploit as mobile nodes.
  • A vehicular cloud is formed by cooperative vehicles using V2V communications, while a roadside cloud links nearby vehicles to local servers through RSUs.
  • A central cloud uses dedicated Internet or vehicular-network servers and supports complicated computation, massive data storage, and global decisions.
  • Vehicular-cloud customization is handled either by GVCC, where a controller schedules virtualized resources, or SVCC, where vehicles select and manage candidate cloud sites directly.
  • GVCC achieves higher resource utilization through global scheduling, whereas SVCC may reduce system overhead by avoiding controller operation.

B. Roadside Cloud

The roadside-cloud design combines local servers with RSUs to provide transient, nearby cloud services. Customized virtual machines use a VM-base and vehicle-specific VM-overlay, then migrate between roadside cloudlets as vehicles move.

  • A roadside cloud combines dedicated local servers, which virtualize resources, with RSUs, which provide vehicle access within radio coverage.
  • A roadside cloudlet is a small roadside cloud site that offers passing vehicles transient cloud services.A vehicle customizes a transient cloud from a nearby cloudlet and replaces it after leaving the serving RSU’s range.
  • The customized VM combines a cloudlet-hosted VM-base containing the basic structure with a vehicle-hosted VM-overlay containing specific resource requirements.
  • During movement between RSUs, the customized VM must be synchronously transferred between roadside cloudlets to maintain service continuity.The paper refers to this transfer process as VM migration.
  • The architecture is intended to support applications that exploit vehicular, roadside, and central cloud resources.

A. Realtime Navigation with Computation Resources Sharing

Cloud-based vehicular networks support shared computation, storage, and bandwidth resources for applications such as realtime navigation, video surveillance, and cooperative downloading.

  • A. Realtime Navigation with Computation Resources Sharing: Realtime navigation combines central-cloud traffic data mining with roadside-cloud messaging to update routes and traffic conditions during travel.A central VM cluster recommends routes, while a roadside VM pushes updates and migrates between roadside cloud sites as the vehicle moves.
  • B. Video Surveillance With Storage Resource Sharing: Cloud-based storage distributes bus video across roadside cloudlets as the bus moves, rather than relying on one large onboard hard drive.The bus uploads video to a roadside VM, which migrates between cloudlets and leaves segments stored along the route.
  • C. Cooperative Downloading and Uploading: Cooperative downloading uses neighboring vehicles and guest VMs to download a large file in parallel.Vehicle A forms a vehicular cloud with vehicles B and C, whose VMs participate in the download.

IV. GAME-THEORETICAL APPROACH FOR RESOURCE ALLOCATION

The paper formulates roadside-cloud VM resource allocation as a non-cooperative game in which VMs compete for computation and storage resources under efficiency, QoS, and fairness considerations.

  • Resource Allocation Requirements: Roadside-cloud resource allocation must efficiently use limited resources while providing sufficient resources for each VM’s task.The allocation problem concerns multiple VMs competing for resources at a cloud site that serves several vehicles simultaneously.
  • Game Model: Each VM requests computation and storage resources, and the cloud allocates available resources in proportion to requested amounts.The model represents VMs as selfish players competing for the cloud site’s total computation and storage resources.
  • Game Model: The VM payoff weights computation and storage preferences while pricing factors discourage resource waste from excessive competition.Virtual Resource Counters record cumulative computation and storage applications for fairness, and parameters are selected for mobile-cloud applications.

B. Nash Equilibrium

The proposed allocation game uses concave VM utility functions to establish Nash equilibrium, with sufficient parameter conditions for uniqueness and a numerical example showing rapid convergence.

  • B. Nash Equilibrium: Concavity of U(ci, mi) with respect to computation and storage resources proves the existence of a Nash equilibrium.The second derivatives with respect to ci and mi are negative under the model.
  • B. Nash Equilibrium: The best response gives each VM’s optimal strategy for fixed applications by the other VMs.The best response is defined as the strategy maximizing the VM’s utility in each iteration.
  • B. Nash Equilibrium: Uniqueness is guaranteed when ∀i, αi ≥4(N−1)λi and βi ≥4(N−1)γi.These inequalities are sufficient conditions obtained by validating positivity, monotonicity, and scalability of the best-response function.
  • B. Nash Equilibrium: Nearly 10 iterations yield equilibrium in a three-VM example, allocating computation as 21.4, 14.3, and 14.3 units and storage as 31.1, 37.8, and 31.1 units.The example uses 50 computation units and 100 storage units, with allocations reflecting differing VM demands.

V. RESOURCE RESERVATION SCHEME FOR VIRTUAL MACHINE MIGRATION

Vehicle mobility creates several VM migration scenarios across roadside cloudlets, vehicular clouds, and central clouds, while movement within one cloudlet requires only radio handoff.

  • Migration Scenarios: Inter-cloudlet movement transfers a guest VM from the old roadside cloudlet to the new one so service can resume.This occurs when the vehicle moves between RSUs connected to different cloudlets.
  • Migration Scenarios: Intra-cloudlet movement requires no VM migration, although radio handoff may temporarily suspend vehicle–VM interaction.Both RSUs connect to the same roadside cloudlet in this case.
  • Migration Scenarios: If vehicle movement disconnects a client from the roadside cloud, its VM can migrate to a neighboring vehicle’s cloud or the central cloud.The destination depends on whether a direct link exists between the relevant vehicles.

B. Resource Reservation Scheme

The resource reservation scheme separates resources for migrated VMs from common resources used by local VMs, coordinating allocation during arrivals and departures. Migration is approved only when re-allocation satisfies all VM resource requests.

  • Resource categories: Reserved and common resources are maintained separately to manage computation and storage during VM migration.Reserved resources are denoted Cr and Mr; common resources are Cc = C − Cr and Mc = M − Mr.
  • Local VM events: Local VMs can use only common resources, and a new local VM is admitted when allocation satisfies all existing VM requirements.Otherwise, the local VM is blocked.
  • VM departures: Resource allocation is performed when local or migrated VMs depart, including when service ends or a VM moves to another cloud site.
  • Migrated VM arrival: A migrated VM shares all available resources with existing local VMs, and migration is approved only if all resource requests remain satisfied.Otherwise, the migration request is rejected.
  • Reserved-resource protection: When no migrated VMs remain at a cloud site, only common resources are allocated while reserved resources are conserved for future migrations.

C. Optimal Resource Reservation

The paper models local and migrated VM dynamics with a continuous-time Markov process, derives blocking and dropping rates, and optimizes reserved resources under a blocking constraint. Simulations compare dropping behavior with and without reservation.

  • System model: VM arrivals and departures are modeled as Poisson processes, yielding a continuous-time Markov process for system-state transitions.
  • System model: The system state records local and migrated VM counts by class, and steady-state probabilities are derived from a 2K-dimensional Markov-chain model.
  • Blocking and dropping: Local VMs are blocked when their total required resources exceed common resources, whereas migrated VMs are dropped when total requirements exceed all available resources.
  • Optimization: The optimal numbers of reserved computation and storage resources are obtained by solving an optimization problem constrained by the blocking rate.
  • Simulation: The simulation uses 50 computation units and 100 storage units, two VM classes, local arrival rates from 0.1 to 0.3, and migrated-VM arrival and departure rates of 0.05 and 0.1.
  • Simulation: The dropping rate of migrated VMs is significantly reduced with resource reservation, demonstrating the proposed mechanism's efficiency.

VI. CONCLUSIONS

The conclusion summarizes a hierarchical cloud-based vehicular-network architecture and resource-management framework. It addresses VM migration under vehicle mobility using resource reservation and reports reduced service dropping during migration.

  • VI. CONCLUSIONS: The architecture enables vehicles to share computational, storage, and bandwidth resources through vehicular, roadside, and central cloud layers.
  • VI. CONCLUSIONS: Cloud resource competition among virtual machines is formulated and solved in a game-theoretical framework.
  • VI. CONCLUSIONS: Virtual resource migration caused by vehicle mobility is addressed using a resource reservation scheme.
  • VI. CONCLUSIONS: Illustrative results indicate a significant reduction in service dropping during virtual machine migration.

ACKNOWLEDAGEMENT

The research was partially supported by national, institutional, Norwegian, European Commission, and COST Action funding programs.

  • Funding: The study received partial support from National Natural Science Foundation of China grants U1035001, U1201253, and 61203117.
  • Funding: Additional support came from the Opening Project of the Key Lab. of Cognitive Radio and Information Processing at GUET, Ministry of Education.
  • Funding: The acknowledgements also list the Research Council of Norway, European Commission FP7 Project EVANS, and European Commission COST Actions IC0902, IC0905, and IC1004.
Loading 1308.6208v1…