Resource Management in Cloud Networking Using Economic Analysis and Pricing Models: A Survey

Nguyen Cong Luong; Ping Wang; Dusit Niyato; Wen Yonggang; Zhu Han

Source-linked AI summary

Resource Management in Cloud Networking Using Economic Analysis and Pricing Models: A Survey

Nguyen Cong Luong, Ping Wang, Dusit Niyato, Wen Yonggang, Zhu Han

arXiv:1701.01963v1Jan 8, 2017cs.GTcs.DCcs.NI

TL;DR

Cloud networking resource management spans allocation, pricing, incentives, and provisioning challenges requiring adaptive approaches. This survey organizes and analyzes economic and pricing models across major cloud-networking systems, comparing their objectives, applications, and open issues.

Problem
Cloud networking must manage resource allocation, bandwidth reservation, request allocation, and workload allocation while pursuing profit, cost reduction, flexibility, and efficient provisioning.
Method
The paper surveys economic and pricing theories, related algorithms, protocols, and incentive mechanisms across data centers, mobile clouds, edge computing, cloud-based VoD, and cloud-based SDWN.
Results
The survey classifies and compares approaches according to cloud-networking models and their issues, highlighting their advantages, disadvantages, applications, and performance objectives.
Takeaways & Limitations
Economic and pricing models provide a framework for designing resource-management mechanisms targeting social welfare, fairness, truthfulness, profit, user satisfaction, and resource utilization.
Takeaways & Limitations
Cost-based pricing does not account for external market factors such as competitors’ pricing strategies or buyers’ perceived value and willingness to pay.

Abstract

from arXiv · show

This paper presents a comprehensive literature review on applications of economic and pricing models for resource management in cloud networking. To achieve sustainable profit advantage, cost reduction, and flexibility in provisioning of cloud resources, resource management in cloud networking requires adaptive and robust designs to address many issues, e.g., resource allocation, bandwidth reservation, request allocation, and workload allocation. Economic and pricing models have received a lot of attention as they can lead to desirable performance in terms of social welfare, fairness, truthfulness, profit, user satisfaction, and resource utilization. This paper reviews applications of the economic and pricing models to develop adaptive algorithms and protocols for resource management in cloud networking. Besides, we survey a variety of incentive mechanisms using the pricing strategies in sharing resources in edge computing. In addition, we consider using pricing models in cloud-based Software Defined Wireless Networking (cloud-based SDWN). Finally, we highlight important challenges, open issues and future research directions of applying economic and pricing models to cloud networking

I. INTRODUCTION

Cloud networking integrates distributed computing and network resources, creating complex, dynamic resource-management challenges. This survey examines economic and pricing models as tools for balancing provider, user, and stakeholder objectives.

I. INTRODUCTION: Cloud networking extends virtualization beyond data centers to provision on-demand cloud and network resources from distributed data centers.
I. INTRODUCTION: Joint management of cloud and network resources requires integrated topology and resource awareness, effective placement, and frequent reconfiguration.
I. INTRODUCTION: Economic and pricing approaches complement system optimization by incorporating profit, cost, and revenue into resource-management decisions.
I. INTRODUCTION: Pricing models address conflicting objectives among users, providers, brokers, and operators through negotiation, demand elasticity, and price discrimination.
I. INTRODUCTION: The survey reviews economic and pricing applications across cloud data centers, mobile cloud networking, edge computing, VoD, and cloud-based SDWN.

1) Cloud data center networking:

The survey covers cloud data center, mobile cloud, and related networking architectures that pool, virtualize, and interconnect resources across domains. These architectures support elastic, metered, multi-tenant, and on-demand services.

1) Cloud data center networking:: Cloud data center networking interconnects servers, storage, gateways, and data centers through intra- and inter-data-center networks.
1) Cloud data center networking:: Virtual machines can migrate within or across providers’ data centers, while network slicing isolates tenants and supports virtual network performance.
1) Cloud data center networking:: Federated cloud networking lets providers borrow overloaded resources or rent out unused resources through outsourcing and insourcing.
1) Cloud data center networking:: Mobile cloud networking integrates cloud computing and network-function virtualization to provision unified mobile network, computing, and storage services.
1) Cloud data center networking:: Mobile cloud networking provides elasticity, service metering, multi-tenancy, on-demand provisioning, resource pooling, and virtualized RAN functions.

3) Edge computing:

Edge computing moves applications, data, and services toward network peripheries, while cloud-based networking architectures and pricing models support programmable and resource-aware management.

3) Edge computing:: Edge computing pushes computing applications, data, and services from central data centers toward network edges.
3) Edge computing:: Locating resources near users reduces traffic, cost, and latency while improving QoS, scalability, reliability, and automation.
3) Edge computing:: Edge computing also alleviates centralized bottlenecks and potential single points of failure and enhances security through edge-oriented data movement.
3) Edge computing:: SDN decouples control and data planes, centralizes control logic in controllers, and makes networks programmable, partitionable, and virtualizable.
3) Edge computing:: The survey classifies pricing approaches as market-based, game-theoretic and auction-based, or Network Utility Maximization based.

2) Differential pricing:

The survey contrasts pricing models that use internal costs, user-specific willingness to pay, provider profit objectives, or social-welfare constraints. Their applications depend on demand information and market conditions.

2) Differential pricing:: Differential pricing charges users differently according to demand and willingness to pay, targeting provider profit.
3) Profit maximization:: Profit maximization selects resource quantity and price to produce the provider’s highest profit, using a demand curve to derive the optimal price.
4) Ramsey pricing:: Ramsey pricing sets different market prices to maximize social welfare subject to a provider-profit constraint, with prices related to demand elasticity.

B. Game Theory and Auction Based Pricing

Game-theoretic pricing models represent strategic interactions among cloud-networking participants with potentially conflicting objectives. Non-cooperative and Stackelberg games provide alternative solution concepts for pricing and resource-management decisions.

Game-theoretic foundations: Game theory models cloud providers, service providers, tenants, and users as participants whose decisions affect one another’s payoffs.Players choose strategies, and payoffs depend on both their own actions and those of other players.
Non-cooperative game: Non-cooperative games model selfish players who maximize individual payoffs without forming coalitions or considering network social welfare.Cloud-resource sellers can compete through pricing strategies in this setting.
Non-cooperative game: A Nash equilibrium is a strategy profile where no player can improve its payoff by changing only its own strategy.The equilibrium is stable because unilateral deviations make payoffs worse, when such an equilibrium exists.
Non-cooperative game: Nash-equilibrium pricing requires checking existence and uniqueness because some games have no equilibrium or multiple equilibria.Multiple equilibria can leave players uncertain about which outcome to select.
Stackelberg game: Stackelberg games model sequential pricing in which a leader moves first and a follower responds after observing the leader’s strategy.The leader’s optimal strategy incorporates the follower’s rational reaction.
Stackelberg game: The Stackelberg leader receives a payoff at least as high as under the corresponding Nash equilibrium, reflecting a first-mover advantage.Applications include bandwidth allocation, access control, provider revenue, client utility, and end-user QoS.

3) Bargaining game:

Bargaining and auction mechanisms determine prices and allocations through negotiation or bidding among cloud-networking participants. The surveyed mechanisms emphasize mutually acceptable exchange, truthfulness, social welfare, and efficient resource matching.

Bargaining game: In bargaining games, sellers and buyers negotiate a bandwidth price that both find acceptable for a successful transaction.The disagreement point contains the seller’s minimum acceptable price and the buyer’s maximum willingness to pay.
Bargaining game: The Nash bargaining solution uses best-response offers and sets the transaction price between the seller’s and buyer’s offers.When k = 1/2, the price splits the difference between the two offers.
Auction: Auctions allocate commodities and establish prices through bidding, with forward, reverse, and double auctions differing in buyer-seller structure.A double auction has multiple buyers and sellers submitting bids and asks to an auctioneer.
Auction: A Vickrey auction charges the winning buyer the second-highest bid, encouraging truthful bidding and strategy-proofness.Truthfulness matters because non-truthful auctions may be vulnerable to manipulation and poor outcomes.
Auction: A VCG auction generalizes Vickrey auctions to multiple commodities, allocating them socially optimally and charging winners for the social-value loss they impose.The payment equals the loss in attainable welfare suffered by remaining bidders.
Auction: Double auctions match buyers’ bids with sellers’ asks, determine clearing prices, and can provide individual rationality, balanced budget, truthfulness, and economic efficiency.The mechanism sorts bids and asks, matches compatible offers, and may set equilibrium prices from buyer and seller prices.
Auction: The Vickrey auction is used more frequently than other surveyed auction mechanisms because of its privacy and truthfulness guarantees.The survey summarizes auction features and suitable cloud-networking scenarios in Table III.

5) Posted-price mechanism:

Posted-price mechanisms assign arriving sellers a fixed price that they may accept or reject, after which sellers can present take-it-or-leave-it offers to buyers.

Posted-price mechanism: A posted-price mechanism gives each sequentially arriving seller a specific take-or-leave price.The seller typically accepts when its actual cost is below the offered price.

C. Network Utility Maximization (NUM)-based pricing

NUM-based pricing incorporates tenant utility and provider cost into cloud-resource allocation through iterative price adjustment. Under the stated convexity assumptions, the process converges to a unique optimal allocation, though concavity may fail for delay-sensitive services.

NUM formulation: Modified NUM maximizes social welfare while accounting for cloud-tenant utility and the cloud provider’s total cost.The setting includes tenants reserving network bandwidth from a provider to deliver video services.
NUM formulation: Each tenant selects resource units by maximizing utility minus the price paid for those units.The tenant response is xi(pi) = arg max xi∈[ai,bi](Ui(xi) − pixi).
Pricing iteration: The provider iteratively updates prices using the difference between provider-side resource choices and tenant requests.The update uses a step size γ and repeats until the allocation vector converges.
Pricing iteration: The allocation converges to a unique optimum because the formulated problem is convex.The paper states that x∗ is unique under this convex optimization formulation.
Limitations: NUM-based resource allocation is challenged when delay-sensitive services have inelastic, non-concave utility functions.The stated optimal solution is feasible only when cloud-tenant utility functions are concave.
Cloud-networking applications: Cloud-networking resource management covers bandwidth, request, and workflow allocation under fluctuating resources, demands, budgets, latency, and cost objectives.Market-based approaches are described for resource utilization, load balancing, task assignment, and completion time.

A. Bandwidth Allocation

The survey reviews pricing and game-theoretic mechanisms for cloud bandwidth reservation and allocation, targeting efficiency, fairness, provider revenue, and user satisfaction. Approaches include auctions, bargaining, differential pricing, and congestion-aware pricing, with trade-offs involving computational tractability and uncertainty.

Bandwidth reservation can lower advance prices but may create oversubscription or undersubscription.
1) VCG auction: VCG auctions target truthful bidding and optimal social welfare for bandwidth reservation, but become computationally intractable for NP-hard allocation problems.
3) Sealed-bid uniform price auction: A two-tier pricing model jointly allocates bandwidth and maximizes provider revenue through premium reservation pricing followed by a sealed-bid uniform price auction.The second tier allocates unreserved bandwidth at a market-clearing price, but does not model future demand or utilization uncertainty.
4) Bargaining game: Nash bargaining uses iterative VM-pair/server negotiation to achieve a unique rate allocation that is Pareto optimal and fair under stated convexity and linear-constraint assumptions.
6) Differential pricing: Differential pricing increases allocated bandwidth up to 28% and provider revenue up to 12% versus deterministic allocation, although revenue gains become slight with larger discounts.
7) Smart data pricing: Smart data pricing adapts prices to congestion to balance minimum bandwidth guarantees and utilization, but its unit-price rule does not ensure an optimal price.

8) Other pricing models:

The survey covers additional pricing and economic models for request, bandwidth, and workload allocation across cloud networking settings. These models pursue utilization, profit, social welfare, cost reduction, SLA satisfaction, and elasticity, while retaining limitations such as forecasting and convergence requirements.

8) Other pricing models: Dominant resource pricing reduces tenant payments by 70-80% versus completion-time pricing, but leaves the bandwidth baseband threshold unspecified.
8) Other pricing models: Ramsey pricing regulates bandwidth demand by maximizing tenant-provider social welfare subject to a preset provider-profit threshold.The model requires the profit threshold and expected demand for each service class.
8) Other pricing models: Dynamic-pricing approaches address workload fluctuations through elastic bandwidth adaptation, but more advanced techniques are needed to forecast those fluctuations accurately.Higher elasticity can increase accumulated service-provider revenue because more demands are met at higher prices.
8) Other pricing models: Economic request-allocation models replace closest-data-center routing to avoid peak-time overload and jointly consider user and provider benefits.
3) Non-cooperative game: A penalty-function extension of a competition game improves social welfare by 10% to 20% over the original competition-game outcome.
2) Spot instance pricing: Spot-instance and on-demand pricing can be combined for workload scheduling that reduces execution cost while guaranteeing workflow deadlines.
8) Other pricing models: Federated cloud networking distributes resources across providers to support capacity, availability, resilience, and multicloud service-level agreements.

1) Profit maximization:

Economic and pricing approaches allocate cloud-network resources while balancing provider profit, user demand, costs, and federation constraints. Reviewed applications span request allocation, price setting, workload prediction, service placement, and bandwidth reservation.

1) Profit maximization:: Profit-maximization models allocate federated-cloud requests among local hosting, outsourcing, and insourcing to improve provider outcomes.The approach partitions requests into subsets hosted locally or outsourced and selects insourcing requests from other providers.
1) Profit maximization:: 42% higher profits and 48% higher request acceptance were reported for some providers versus operation without federation.These simulation results compare the proposed federated approach with the case without federation.
1) Profit maximization:: Price bounds use provider costs and competing insourcing prices, then feed each provider’s revenue-maximization problem.Lower bounds protect revenue over overall cost, while upper bounds reflect competing providers’ prices.
1) Profit maximization:: Workload prediction and differential pricing account for users’ future resource consumption and relinquishing probabilities.The proposed resource price is proportional to a user’s service-relinquishing probability.
1) Profit maximization:: Cost-based service placement in federated hybrid clouds minimizes users’ total fixed and variable costs, though latency and other performance factors remain relevant.The reviewed model includes hardware, software, electricity, connectivity, and inter-cloud transfer costs, while noting the need to consider service latency for SLAs.
1) Profit maximization:: Double-auction bandwidth reservation selects sellers and buyers subject to bandwidth sufficiency and ex-post budget balance.The auctioneer requires total buyer charges to be less than total seller payments, ensuring non-negative auctioneer profit.

2) Conventional auction:

Conventional auctions support resource allocation across mobile cloud, Cloud-RAN, and edge-computing settings. The surveyed mechanisms trade communication overhead, computational complexity, and information requirements against utility, efficiency, pricing, and participation goals.

2) Conventional auction:: Dutch-auction bandwidth redistribution reaches a Nash equilibrium maximizing gateways’ utilities, but assumes gateways know others’ bids.The bid-information assumption is identified as unrealistic.
2) Conventional auction:: English-auction sharing redirects users from overloaded gateways to neighboring gateways using location and QoS information.The mechanism addresses overload at one gateway by sharing users with neighbors.
B. Resource management in Cloud-RAN: Cloud-RAN pricing coefficients regulate RRH fronthaul capacity while optimization minimizes transmission power under capacity and QoS constraints.Binary search adjusts coefficients, while the second problem is solved with a gradient method.
B. Resource management in Cloud-RAN: Allowing more RRHs to serve one user reduces total transmission power but increases computational complexity.This trade-off is reported from simulation results.
2) Conventional auction:: Auction-based approaches are well suited to MCN bandwidth allocation, whereas Cloud-RAN pricing applications remain relatively few.The survey identifies further research as necessary for extending Cloud-RAN pricing results.
2) Conventional auction:: Group-buying auctions lower the final service price as more winners participate, stimulating users to choose cloudlet-group services.The auction price curve is obtained by maximizing the cloudlet group’s expected profit.
2) Conventional auction:: Truthful bidding improves buyer utility, while final matching efficiency reaches only around 50% of the optimal strategy.This result concerns the auction model using second-highest-bid payment.

2) Non-cooperative game:

Non-cooperative and profit-oriented pricing models distribute edge, volunteer, and external network resources among participants with differing utilities, budgets, costs, and demand conditions.

2) Non-cooperative game:: Broker-based cloudlet allocation assigns reserved computation and bandwidth resources from cloudlets and public clouds to mobile users.Long-term reservation and on-demand requests apply at public clouds, while cloudlets use bid-proportion allocation because resources are limited.
2) Non-cooperative game:: Mobile users exchange cloudlet CPU, storage, and broadband resources through supply-and-demand pricing, maximizing individual payoff under budget constraints.Each user may act as both buyer and seller during mobility, and the convex problem is solved by a primal-dual algorithm.
2) Non-cooperative game:: Mobile telecom cloud brokers use provider discounts and optimization to offer cloud services within feasible offer-price and cost conditions.Linear programming with rounding or a min-cost greedy method determines the brokerage solution.
2) Non-cooperative game:: Volunteer-computing brokers earn significantly higher total profit under low demand than high demand because high-demand prices must remain acceptable to users.The broker cannot raise user prices freely when demand is high.
2) Non-cooperative game:: Demand-based pricing charges users according to resource demand to support resource utilization and low costs.The approach addresses users who prefer payments proportional to QoS.
2) Non-cooperative game:: VCG bandwidth allocation yields significantly higher social welfare than first-price and Vickrey auctions as the probability of user lying varies.The mechanism also seeks provider revenue maximization while allowing users to misreport priority.

1) Client-assisted cloud storage system:

Client-assisted and socially organized edge-cloud systems use auctions, posted prices, incentives, and reputation to pool user resources and allocate tasks or storage. These approaches address distributed participation, accountability, latency, and resource scarcity.

1) Client-assisted cloud storage system:: Online reverse auctions build storage pools from underutilized user storage and bandwidth despite asynchronous arrivals of users and providers.Users act as sellers and storage service providers procure their resources.
1) Client-assisted cloud storage system:: Client-assisted storage can introduce request latency because users must communicate with the service provider when another user needs resources.Distributed cloud models are presented as an alternative for considering interactions between demand and supply users.
1) Client-assisted cloud storage system:: Self-organization clouds use reverse auctions and resource discovery to match buyers with candidate sellers in a P2P overlay.Kademlia-based discovery precedes reverse-auction execution.
1) Client-assisted cloud storage system:: Non-money incentives reduce buyer costs as seller numbers increase while preserving fair and stable resource allocation, but reverse auctions fail when resources are insufficient.The incentive may represent resources received in the future, with coupons or reputation scores as alternatives.
1) Client-assisted cloud storage system:: Social-cloud posted prices manage sequential service offers using user identity and credit balances maintained by a bank.The framework establishes accountability through social relationships and financial records.
1) Client-assisted cloud storage system:: A social-cloud reverse Vickrey auction selects the lowest-asking seller and creates an SLA with bank-mediated credit transfer.The payment policy is intended to prevent provision-user misreports.
1) Client-assisted cloud storage system:: Crowdsensing reverse auctions assign sensing tasks to phone users according to marginal value contributions, while mobility and malicious data motivate tracking and reputation mechanisms.Selected users receive payments not less than their asks, determined using their marginal contributions.

A. Cloud-Based VoD Models

Cloud-based VoD studies apply auctions, broker pricing, games, and distributed updates to allocate bandwidth and maximize welfare or provider objectives. These approaches reduce coordination overhead, establish equilibria, and address bandwidth costs, while broader multi-provider markets remain open.

A. Cloud-Based VoD Models: Cloud-based VoD bandwidth allocation uses competition-based pricing, including combinatorial auctions, to match video-demand groups with cloud-provider capacity.The surveyed models include one VoD provider with multiple cloud providers, as well as multiple tenants and cloud providers coordinated by a broker.
A. Cloud-Based VoD Models: A broker can save more than 30% of bandwidth reservation cost on average compared with tenants reserving bandwidth individually.The broker sets price bounds and directs tenant demands to cloud providers; tenant demand is assumed to follow a Gaussian distribution.
A. Cloud-Based VoD Models: Tenant competition in a free market converges to a unique Nash equilibrium, but competition among brokers may reduce broker profit to zero.The equilibrium persists with multiple brokers because the game is played among tenants.
A. Cloud-Based VoD Models: Local-information pricing updates reduce message-passing overhead and require fewer convergence iterations than the cited comparison methods.The proposed approach in converges in an average of 10 iterations, versus 100 for primal gradient descent and 50 for consistency pricing.
A. Cloud-Based VoD Models: P2P-assisted VoD pricing incentivizes users to obtain video from peers and cache content, targeting lower cloud bandwidth costs and improved scalability.The model requires incentives for peer downloading and for using peer memory and upload bandwidth.

1) Stackelberg game:

Stackelberg games and related pricing mechanisms coordinate resource sharing among providers, users, peers, and servers in cloud-assisted streaming and multimedia networks. The surveyed approaches optimize utilities, budgets, caching, and bandwidth incentives, while information asymmetry can prevent equilibrium attainment.

1) Stackelberg game:: A usage-based Stackelberg pricing scheme lowers cloud bandwidth consumption by encouraging users to download video chunks from peers.The provider estimates bandwidth usage, sets service prices, and users select bit rates to maximize satisfaction minus price.
1) Stackelberg game:: Double auctions support decentralized trading of surplus peer upload bandwidth, with separate sub-markets for individual video segments.Peers may act as buyers or sellers in the resource-sharing market.
1) Stackelberg game:: Reward prices encourage peers to cache popular video replicas, with prices increasing with popularity and required replication.The price is also inversely proportional to storage probability.
1) Stackelberg game:: Budgeted Stackelberg models prove a unique Nash equilibrium for helper strategies and a unique utility-maximizing server budget.The server’s utility is the gain from transmitted video minus helper rewards, characterized using a two-parameter rate-distortion model.
1) Stackelberg game:: Reverse auctions address missing helper-utility information by selecting bandwidth suppliers within a budget, although winner determination is NP-hard.The server’s utility is submodular, enabling budget-feasible mechanism-design methods for winner determination and payments.
1) Stackelberg game:: A cheat-proof bandwidth strategy makes mobile users bid their minimum requirements to obtain their desired bandwidth, unlike the non-cheat-proof method.Under the non-cheat-proof method, users can receive the same bandwidth while paying less.
1) Stackelberg game:: Nash equilibrium cannot be obtained when desktop users lack information about one another’s bandwidth and pricing strategies.Learning-based algorithms are proposed to update strategies toward increasing desktop-user utility.

VIII. APPLICATIONS OF ECONOMIC AND PRICING MODELS FOR RESOURCE MANAGEMENT IN CLOUD-BASED SDWN

Cloud-based SDWN combines centralized SDN control with cloud data-center resources to manage wireless-network resources using economic and pricing models. The surveyed approaches address cost, bandwidth, offloading, sharing, and incentive-alignment objectives.

Architecture: Cloud-based SDWN uses an SDN controller to monitor and allocate data-center resources to users through cellular networks and WiFi hotspots.This combines data-center networking with centralized SDWN control for complex network management.
Resource allocation: Economic and pricing models support centralized bandwidth allocation while maximizing the payoffs of cloud providers, service providers, and mobile users.The models allocate bandwidth from cloud providers to service providers and mobile users.
Resource sharing: The proposed cooperative Nash bargaining approach increased allocated buyer resources and seller revenue compared with buyer competition, although competition better reflects resource scarcity.The objective was to guarantee user QoS while increasing the total utility of sellers and buyers.
Mobile data offloading: Contract models trade access-point offloaded traffic for payments, with optimal contracts balancing the base station’s gain against access-point participation constraints.Perfect discrimination maximizes base-station payoff but gives access points zero payoff; incentive-compatible variants address this issue.
Mobile data offloading: The proposed home-network pricing approach improved both user and service-provider payoffs by 400% over best-effort usage-based pricing.The passage identifies limited backhaul capacity as a future consideration.

IX. REVIEW SUMMARY, OPEN ISSUES AND FUTURE

The survey concludes that economic and pricing models address cloud-networking resource-management problems beyond traditional optimization, while identifying mechanism, forecasting, and systems challenges. It highlights open issues involving strategic behavior, workload prediction, multipath routing, cloud robotics, and NFV/SDN integration.

Review summary: Economic and pricing models address cloud-networking issues where traditional algorithms become less effective or cannot be applied.The survey frames economic aspects as relevant to both business development and system design and optimization.
Open issues: False-name bidding can invalidate dominant-strategy incentive compatibility in double auctions, motivating robust mechanisms against fictitious identities.A bidder may submit multiple bids under different identities to gain additional profit.
Open issues: Auction bidders may collude to suppress competition and prices, degrading the efficiency of cloud-resource allocation.The issue is reported for VCG, combinatorial, and double auctions.
Future directions: Only a Markov chain model was identified for cloud resource-demand forecasting, motivating advanced methods such as fuzzy logic, neural networks, and machine learning.Demand fluctuation affects resource availability, pricing policy, and cloud-provider profit.
Cloud data-center networking: Pricing-based multipath routing can select among alternative data-center paths using asking prices to support load balancing and reduce network latency.In reverse auction routing, the path with the lowest asking price is selected for forwarding.
Emerging systems: NFV and SDN support software-based network-service design and management, with NFV decomposing network functions from physical equipment and potentially reducing CAPEX and OPEX.The survey also describes cloud robotics as integrating robots with cloud computing and networking resources.