Review and Analysis of Networking Challenges in Cloud Computing

• Cloud performance depends heavily on networking, so infrastructure limitations can impair data-intensive and high-performance applications and motivate changes to network functionality.Emerging scenarios include dynamic loads, data mobility, data-based addressing or routing, heterogeneous resources, federation, and energy efficiency.Source: §1 · pp. 1–2
• The survey examines how network architectures, protocols, and algorithms should evolve to support cloud services in dynamic, resource-constrained environments.It considers reliable and efficient networking-resource allocation, virtualization, and emerging security aspects.Source: §1 · pp. 2–3
• The article reviews major network functionalities to modify or tune, surveys networking challenges, and identifies open issues that could hinder future cloud-service deployment and management.Its discussion also considers research directions for future networks through IoT and NFV scenarios, and current and future cloud security challenges.Source: §1 · p. 3
• The paper progresses from cloud fundamentals and technologies to networking challenges, future research directions, security, selected research challenges, and conclusions.The background section is intended mainly for readers building knowledge of cloud computing; specialized readers may skip to Section 3.Source: §1 · p. 3

2.1 History and Emergence of Cloud Computing

• The term “cloud” abstracts underlying computing, networking, and storage infrastructure, while Table II traces cloud computing’s historical evolution from the 1960s to 2011.Source: §2.1 · p. 3, Table II · p. 4
• Big Data’s growth is tied to cloud support for storage, scalable distributed processing, collaboration, and secure, private mobile access; IoT and NFV also raise cloud network challenges.Source: §2.1 · pp. 3–4

2.2 Definition of Cloud Computing

• Cloud computing provides on-demand access to shared, configurable resources that can be rapidly provisioned with little management effort, without customers needing to manage provisioning details.Source: §2.2 · pp. 4–5

2.3 Foundations of Cloud Computing

• Cloud computing combines hardware, distributed computing, Internet services, and network management; its foundations include virtualized computing and networking, ubiquitous access, self-service, and automated resource management.Virtualization can improve security, reliability, compatibility, utilization, maintenance, load balancing, and recovery; automation adapts resource use to customer needs and system status.Source: §2.3 · pp. 5–6, Figure 1 · p. 5
• Grid computing distributes complex or time-consuming applications across machines, but diverse operating systems and runtime environments create portability barriers that virtualization may address.GridGain uses MapReduce to split tasks into parallel subtasks and aggregate their results.Source: §2.3 · p. 6

2.4 Cloud Computing Service Models

• The service-model discussion compares how providers and clients divide control, then considers real deployments and each model’s functional strengths and weaknesses.Source: §2.4 · p. 6

Software as a Service (SaaS)

• SaaS providers control most of the software stack, while subscribers are generally limited to controlling the application level.Subscribers cannot configure middleware or the operating system of each virtual machine.Source: Software as a Service (SaaS) · pp. 6–7
• SaaS can compose services from distinct providers into higher-value solutions when a single service does not meet all requirements.The passage describes Salesforce components and Programmable Web APIs as examples of building customized services.Source: Software as a Service (SaaS) · p. 7
• SaaS nearly eliminates customer deployment and maintenance tasks by relying on the provider to perform them (Sridhar_a 2009).Source: Software as a Service (SaaS) · p. 7
• Subscribers cannot add applications, have only limited personalization, and cannot scale allocated resources because providers control delivery monitoring and middleware.PaaS is presented as an alternative for requirements that SaaS does not ensure.Source: Software as a Service (SaaS) · p. 7

Platform as a Service (PaaS)

• PaaS gives customers control over applications and configurations and an API to adjust computational resources dynamically.Examples of adjustable resources include memory and storage disk.Source: Platform as a Service (PaaS) · p. 7
• Vendor-specific languages and APIs make it difficult to move legacy applications into PaaS or between cloud providers.The latter difficulty arises when a customer changes providers.Source: Platform as a Service (PaaS) · p. 8
• These migration problems could be avoided if PaaS vendors agreed on a standard API or customers chose an IaaS provider.Source: Platform as a Service (PaaS) · p. 8

Infrastructure as a Service (IaaS)

• IaaS gives subscribers configurable, isolated virtual machines and access to processing, storage, and networking without management of the underlying physical system.Subscribers can choose a supported guest operating system, install applications, and adjust VM resources; the hypervisor shares hardware across VMs.Source: Infrastructure as a Service (IaaS) · p. 8
• Amazon EC2 offers configurable VM counts and features including elastic storage, monitoring, auto scaling, load balancing, HPC, and VM import/export.These features support demand-based capacity, traffic rerouting from unhealthy instances, scientific and engineering workloads, and workload portability; EC2 is described as enabling failure-resilient applications.Source: Infrastructure as a Service (IaaS) · pp. 8–9
• IaaS challenges include virtual networking, cloud extension, and federation, while simplified APIs can support on-demand resource allocation and overlay-network management.A proof of concept created an IPv6 network across cloud locations around the world (Boutaba, et al. 2014) (Strijkers, et al. 2014).Source: Infrastructure as a Service (IaaS) · p. 9
• The service taxonomy is rapidly expanding beyond SaaS, PaaS, and IaaS to include many “X as a Service” categories, including Network and Everything.Other listed categories include Backend, Business Process, Database, Information, Infrastructure, Storage, Platform, Security, and Software.Source: Infrastructure as a Service (IaaS) · pp. 9–10
• Networking aspects that can impair cloud-system performance have received less attention than cloud functionality, motivating the paper’s next section.The paper identifies this as a novel perspective for further study.Source: Infrastructure as a Service (IaaS) · p. 10

3.1 Reliable Communications

• FCoE supports Ethernet–SAN interconnection, while IEEE 802.1Qbb, 802.1Qau, and 802.1Qaz address frame loss, congestion management, and bandwidth allocation, respectively (ANSI/INCITS 2009; IEEE_a 2011; IEEE_b 2010; IEEE_c 2011).802.1Qbb uses hop-by-hop priority flow control; 802.1Qau signals congestion end-to-end, and 802.1Qaz allocates bandwidth among traffic classes.Source: §3.1 · pp. 10–11
• ConEx lets sources report prior packet congestion, enabling ingress traffic shaping without flow-specific switch configuration, while OpenFlow offers a centralized controller-based alternative (Mathis and Briscoe 2014; Briscoe and Sridharan 2014; McKeown, et al. 2008).The ingress-policing contribution describes per-flow performance through congestion feedback; OpenFlow instead configures flows in switches to establish the data path.Source: §3.1 · p. 11

3.2 Efficient Communications

• SPB activates multiple paths to balance data-center traffic and avoid spanning-tree bottlenecks, while TRILL routes encapsulated frames over Layer 3 techniques and supports large flat Layer 2 clouds (IEEE_d 2012; Eastlake, et al. 2011; Perlman and Eastlake 2011).SPB supports up to 16 million VLANs, compared with the traditional limit of 4,096, and TRILL adds an outer header to frames.Source: §3.2 · pp. 11–12
• OSPF extensions can support scalable data-center routing, while BGP/MPLS can benefit from per-flow QoS/QoE and multiple exit routes that improve recovery and load balancing (Retana, 2013; Astuto, et al. 2014; Gupta, et al. 2014; Mohapatra, et al. 2015).Multiple exit points can also contain failures within an autonomous system and permit router maintenance with less traffic loss.Source: §3.2 · p. 12

3.3 Virtual Networking

• Virtual switches provide intrahost VM switching and policy controls, but unknown-destination forwarding can create sniffing or spoofing risks, and VM moves can require complex cross-switch VLAN reconfiguration (Wu, et al. 2010; Sridhar_b 2009; Chowdhury and Boutaba 2010; Layland 2010).SDN/OpenFlow separates control from switches, but introduces control-traffic overhead and controller-failure risks; proposed mitigations include optimized signaling, redundant controllers, and hairpin switching (Astuto, et al. 2014; Stallings 2013; IEEE_e 2012).Source: §3.3 · p. 13
• SDN approaches can manage wireless resources and expose cloud-networking primitives, while Shin and Gu (2012) describe diverting suspected traffic to security appliances for inspection.Wireless examples include on-demand virtual access points, scheduling, dynamic spectrum use, and intercell interference coordination.Source: §3.3 · p. 14

3.4 Elastic Allocation, Federation, and Interoperability

• Section 3.4 summarizes additional networking research activities concerning elastic allocation, federation, and interoperability in Table VI.Source: §3.4 · p. 14, Table VI · p. 14

• Seibold et al. (2012) propose that database systems communicate resource requests and adjust usage, with virtualization helping large databases handle peak demand.The cooperative approach uses virtual machines to deploy requested resources.Source: Elastic Allocation of Cloud Resources… · pp. 14–15
• Birke, Chen and Smirni (2012) provide a large-scale data-center workload survey that may support more reliable future planning.Source: Elastic Allocation of Cloud Resources… · p. 15
• Metri et al. (2012) identify accurate load prediction and dynamic virtual-machine provisioning as necessary for data-center energy efficiency.Their findings come from extensive experiments on data-center energy efficiency.Source: Elastic Allocation of Cloud Resources… · p. 15

Cloud Federation

• Single sign-on lets customers authenticate once across cloud services through a distributed third-party authentication server, with Kerberos 2014 supporting this functionality.The server authenticates the customer and supplies credentials to the selected provider.Source: Cloud Federation · p. 15
• Portable VM formats enable migration across cloud technologies; the DMTF’s OVF describes VM status in a vendor-neutral format (DMTF_a 2014).Inter-provider federation may also depend on peering agreements and compatible APIs, potentially preceding formal standards.Source: Cloud Federation · p. 15
• Federation also requires coordination between management and billing systems and new business models, with customers billed for resources or data used at each provider (Sen, et al. 2013).Providers may adapt mobile-operator models for peering or roaming agreements.Source: Cloud Federation · p. 15
• Federation can reduce electricity costs by 15% by directing requests to data centers where electricity is cheaper (Sakamoto, et al. 2012).The proposed management policy sends energy-price requests to federated data centers to help target workloads.Source: Cloud Federation · pp. 15–16

Interoperability

• The DMTF’s OVF supports VM configuration and status portability across physical machines and hypervisors, while its Open Cloud Standards Incubator studies portability, federation, and service adaptation (DMTF_a 2014).Coordination efforts seek to reduce redundancy and incompatibility among cloud specifications.Source: Interoperability · p. 16
• OASIS investigates cloud application interoperability, real-time authorization, cross-cloud data sharing, and functional criteria for clouds used by public administrations (OASIS 2014).Its work treats service-oriented architectures as a basis for cloud computing.Source: Interoperability · p. 16
• SNIA’s Cloud Data Management Interface lets customers attach metadata describing data-service requirements, including backup, archiving, encryption, authentication, and authorization.The Cloud Storage Initiative proposes personalized cloud storage.Source: Interoperability · p. 16

4.1 Reliable Communications

• A recent contribution optimizes resource-allocation and fault-tolerance tradeoffs, while demand is driving 400 Gbps Ethernet development and expectations of eventual 1 Tbps Ethernet (Bodik, et al. 2012; Stallings, 2015).The resource-allocation and fault-tolerance studies cited in the passage include, (Shieh, et al. 2011), (Ballani, et al. 2011), (Duffield, et al. 1999), (Ricci, Alfeld and Lepreau 2003), (Agarwal, et al. 2010), (Amiri, et al. 2000), (Bansal, et al. 2008), and (Yu, Gibbons and Nath 2006).Source: §4.1 · p. 17

4.2 Efficient Communications

• Safe, efficient sharing of unused resources across complex cloud federations remains challenging, and congestion-proportional ConEx is still an open issue (Popa, et al. 2011; Mathis and Briscoe 2014).Cloud resource sharing and ConEx are proposed approaches, but the passage describes the approach as unresolved.Source: §4.2 · p. 17
• Cloud communication metrics need standardization and operational data-center evaluation because the proposed framework’s unexplained results hinder replication (Fiandrino, et al. 2015).Further work is needed to explain how results are obtained so others can reproduce them and build on the findings.Source: §4.2 · p. 17

4.3 Virtual Networking

• Future mobile multimedia could replace client-server Web applications with collaborative peer-to-peer cloud services, alongside infrastructure convergence, cloud offloading, and network-function virtualization managed through promising SDN (Panagiotakis, et al. 2015; Koumaras, et al. 2015).The envisioned cloud-based applications communicate, stream, and transfer adaptive events and content to clients; the heterogeneous virtualized ecosystem creates management and administration challenges.Source: §4.3 · pp. 17–18
• Intelligent, efficient management of networking resources in mobile cloud computing is identified as a research direction (Mastorakis, et al. 2015; Fernando, Loke and Rahayu 2013).Source: §4.3 · p. 18

Cooperation in Cloud Computing

• Cloud participants can be encouraged to cooperate through common goals or monetary, social, and reputation incentives, each with distinct limitations (Huerta-Canepa and Lee 2010; Charilas, et al. 2011; Tanase and Cristea 2011; Hwang, Shin and Yoon 2008).Common goals require shared activity; monetary schemes raise questions about business models, credit, transaction security, pricing, and tariffs; social incentives risk free riding, while reputation values may lack fairness and trust.Source: Cooperation in Cloud Computing · p. 18

Mobile Cloud Computing and Network Functions Virtualization

• Mobile Cloud Computing seeks enhanced QoE through traffic offloading, service migration, and caching data near users to reduce RTT and jitter (Wang, Chen, & Wang 2015).Migration across federated clouds can tailor services to user location, profile, and terminal characteristics; caching may use proxies, access points, base stations, or terminals.Source: Mobile Cloud Computing and Network… · p. 18
• NFV abstracts network functions from hardware into software building blocks that can be chained on standard servers, switches, storage, or cloud infrastructure.Potential applications include virtualized load balancers, firewalls, intrusion detection, WAN accelerators, mobile-device power control (Mavromoustakis, et al. 2015), and new MCC business models.Source: Mobile Cloud Computing and Network… · p. 19
• Migrating NFV to cloud environments is challenging, and mobile-cloud resource management must address coverage, interference, congestion, and battery autonomy.Proposed approaches include offloading, power control, control theory, data mining, machine learning, radio-spectrum management, and MCC business models (Mavromoustakis, et al. 2015; Katzis 2015).Source: Mobile Cloud Computing and Network… · p. 19
• Mobile multimedia streaming needs cross-layer mechanisms and resource and network-fluctuation prediction to mitigate congestion and optimize end-user QoS and QoE (Batalla 2015; Kryftis, et al. 2015).The scenario concerns audiovisual content delivered from media clouds through the Internet to mobile devices.Source: Mobile Cloud Computing and Network… · p. 19
• Federated cloud storage and content-delivery services need abstraction across provider infrastructures, while data location and provider changes raise legal and business concerns (MetaCDN 2014; Akamai 2014; Voorsluys, Broberg and Buyya 2011; SECCRIT, 2014).Physical locations of virtual machines and storage affect applicable national laws on security breaches or tampering; provider ownership changes or closure also affect customer data and applications.Source: Mobile Cloud Computing and Network… · p. 19
• SOA research spans heterogeneous-network virtualization, service discovery, QoS-aware composition, and multi-cloud SDN applications, but wide-area SDN viability remains unproven (Duan, Yan and Vasilakos 2012; Chowdhury and Boutaba 2010; Jain and Paul 2013; Tzanakaki, et al. 2013; Rambold, et al. 2009; Strunk 2010).One proposed wide-area application is support for lambda-path networks using wavelength switches rather than packet switches.Source: Mobile Cloud Computing and Network… · pp. 19–20
• Cloud-based IoT must support real-time analysis of sensor data and historical analysis of previously collected structured or unstructured data.IoT combines wireless access, cloud, and data-analysis APIs, with applications including e-health, engineering, transportation, and social use cases.Source: Mobile Cloud Computing and Network… · p. 20

5. SECURITY ASPECTS

• Cloud security and resilience matter especially as customers place critical services in the cloud, where systems should maintain suitable service despite compromise or outages.The paper says trustworthiness is increasingly important and that security’s functional perspective needs further investigation because distributed-cloud performance depends heavily on underlying networks (SECCRIT, 2014; Sterbenz, et al. 2010).Source: §5 · p. 20

5.1 Background

• Cloud computing’s aggregation of technologies and distributed solutions creates new vulnerabilities and threats of damaging attacks.Source: §5.1 · p. 21

Generic Cloud Security Aspects

• Cloud providers should combine patched infrastructure, VM isolation, authentication, secure connectivity, and controls against unauthorized access and DDoS traffic.The paper also notes patch incompatibility and recommends rollback to a stable configuration; VPN tunnels can secure communication among cloud resources, and detected DDoS traffic can be discarded via a “black hole.”Source: Generic Cloud Security Aspects · pp. 21–22

Security Risks Depend Upon the Cloud Service Model

• SaaS inherits Web-service risks, including interface loopholes, malicious code that breaks isolation, and message tampering by man-in-the-middle attackers.These risks are discussed by Subashini and Kavitha (2011) and Fernandes, et al. (2014).Source: Security Risks Depend Upon the Cloud… · p. 22
• PaaS multitenancy requires isolation between customers, but stronger isolation trades off against resource consumption (Rodero-Merino, et al. 2012).Source: Security Risks Depend Upon the Cloud… · p. 22
• IaaS risks include VM escape, vulnerable hypervisors, and bridged networking that may expose traffic to promiscuous-mode VMs.A successful VM escape can let attackers monitor other VMs, shared resources, and CPU use, or shut down VMs (Grobauer, et al. 2011); bridged networking may allow traffic capture (Pearce, et al. 2013).Source: Security Risks Depend Upon the Cloud… · pp. 22–23

5.2 Future Developments

• Future work includes hybrid cloud IDPS, trust methods for privacy, and technical, legal, and liability guidance for critical-infrastructure cloud use.Signature detection is fast but misses previously unknown attacks, while anomaly detection has opposite functional characteristics; SECCRIT investigates security risks and European legal frameworks, and unclear liability and regulation may limit adoption.Source: §5.2 · pp. 23–24
• Cloud defenses should prioritize simultaneous attacks and adapt to scale, while applying minimally appropriate policies to balance security checks against customer resources.The paper recommends self-managing and self-learning policies, and notes that security solutions must adjust to network-node numbers, node heterogeneity, and traffic load.Source: §5.2 · p. 23

Security Risks of an Emerging Cloud Service Model: Data as a Service

• DaaS users outsource their data to cloud providers, becoming third parties and losing control over the data.Dropbox is given as an example of this usage scenario.Source: Security Risks of an Emerging Cloud… · p. 24
• Symmetric-key encryption is vulnerable when a revoked user rejoins the system.Encrypting data before exporting it to the cloud is described as an initial privacy solution, but the legacy symmetric-key approach has this weakness.Source: Security Risks of an Emerging Cloud… · p. 24
• Samanthula et al. (2015) propose homomorphic encryption and proxy re-encryption to prevent privacy leakage after revoked users rejoin and prevent collusion with the cloud provider.The scheme also supports secure query processing over encrypted data stored across a federation of clouds.Source: Security Risks of an Emerging Cloud… · p. 24

6. OPEN ISSUES

• Dynamic cloud-resource management must balance fault tolerance, energy use, network utilization, load balancing, congestion, and availability while supporting diverse services.The paper notes that SDN can limit packet-flow rates and intelligently forward packets to mitigate congestion and optimize availability.Source: §6 · p. 24
• Cloud security research must address solution harmonization, multitenancy privacy, insider threats, and the balance between security requirements and performance.SDN-based IDPS features may enhance data-center security, while legal research should accompany security and resilience work for critical infrastructure IT.Source: §6 · pp. 24–25
• Outsourcing middleboxes to the cloud could reduce customer-network cost, management complexity, and overhead (Sherry, et al. 2012).Network hypervisors coupled with SDN-controlled vSwitches can provide overlay virtualized networking services independent of underlying hardware, topology, and operator domains (Vmware NSX 2014; Silva, et al. 2013).Source: §6 · p. 25
• Edge admission control can manage accepted mobile flows by Class of Service in resource-constrained settings, with potential relevance to MCC (Silva, et al. 2013).Source: §6 · p. 25
• The proliferation of Internet-connected devices in settings from cities to industry may accelerate IPv6 adoption as devices require IP addresses to communicate with cloud systems.Source: §6 · p. 25
• Cloud stakeholders must address security, privacy, trust, resilience, interoperability, availability, fault tolerance, disaster recovery, resource management, and power efficiency.The paper identifies the “green” aspect of cloud-system power efficiency as an important challenge (Sharkh, et al. 2013).Source: §6 · p. 26

7. CONCLUSION

• Networking concerns tied to outsourcing, virtualization, shared public environments, security breaches, and resilience hamper cloud computing’s fast adoption.Source: §7 · p. 26
• The article reviews cloud computing’s origins, technologies, architectures, offerings, networking support, and industry, academic, and standards contributions, and highlights research needs.Source: §7 · p. 26