With the development and gradual maturation of public clouds, more and more enterprises are starting to use public cloud services, at first to carry non-key services or disaster recovery services and, eventually, to carry key applications. Of course, development does not happen overnight, and, in addition, some applications will ultimately still use private clouds. Therefore, in the long term, hybrid clouds are the fundamental form of the enterprise IT architecture. It is well known that the biggest advantage of cloud services is flexible IT resource sharing, and so networks across private and public clouds must also be flexible for effective hybrid cloud development support.

On-demand Networks are a Fundamental Requirement of Cloud Service Development

Hybrid clouds provide enterprise applications with the benefits from both public and private clouds, but they also face a variety of challenges regarding inter-cloud network interoperation, network latency, multi-vendor heterogeneity, and network resource scalability necessitated by flexible cloud resource scalability.

• Networks require higher performance imposed by scale and demand: As cloud application development continues apace, the number of enterprise users, the scale of DCs, and the amount of cloud service data and traffic, as well as meeting user needs are all continually increasing. Not only is there a need to deal with the issues of mass NE management and dynamic scaling, but also increasingly higher requirements are put forward by a large number of tenants regarding network capacity, service deployment time, and convergence performance.
• Cross-DC unified resource orchestration and VPN deployment between DCs: Interconnections among multiple DCs often require complex external networks spread across long distances. In addition, support for the unified orchestration of private and public cloud services is required so applications can be automatically migrated within public and private clouds based on service requirements.
• Multi-vendor interoperability: Hybrid cloud infrastructure is, by necessity, multi-vendor heterogeneous. Achieving seamless integration of products from different vendors, as well as unified management and O&M, is a problem that all hybrid cloud solutions must tackle.
• E2E streamlined DC, campus, and WAN (IP/optical) networks: As enterprises deploy data and services on the cloud, the distance between users and their data objectively increases. As a result, network borders must be broken in order to achieve the unified management and scheduling of network resources. A unified network resource pool can be achieved through the synergy of IP and optical layers.
• Clouds require application-oriented, on-demand, auto-scaling, and cloud-adaptive networks: Application-oriented networks enable IT administrators to define network requirements using service-based language from service perspectives, as well as implement automated application-based network resource scheduling to allow their networks to support on-demand intelligent cloud connection.

SDN Architecture: Best Choice across Private and Public Clouds

SDN redefines network capabilities from a software perspective, solving problems that many traditional networks simply cannot. SDN technology enables networks to move with the cloud — allowing dynamic changes in real time based on application and service requirements. SDN also improves network O&M efficiency — reducing O&M costs by 40 percent through automated network management and fault rectification. In addition, SDN implements network traffic steering in a centralized control manner, significantly improving network utilization and fault tolerance. For example, with the help of the Huawei Agile Controller, 21Vianet has achieved an increase from 50 to 80 percent in link utilization between DCs.

Through its open network capabilities, SDN allows enterprise users to configure their networks based on service requirements, as well as implement Network-as-a-Service (NaaS) and network and bandwidth on demand. This is the most important part of a hybrid cloud’s network architecture. In addition, to better meet hybrid cloud requirements, SDN networks provide the following functions and features. 

Flexibility and Scalability

The hybrid cloud network is scalable enough to handle sharp increases in cloud tenants and services. When the number of devices in a DC, on a WAN, or on a campus network increases, the SDN controller is still able to maintain unified control of the network. Two key technologies support SDN’s network control flexibility: controller cluster and controller federation.

• Controller cluster: Limited by the server’s processing capabilities, single-node controller networks can control a maximum of thousands of hosts. However, using a cluster consisting of multiple controller nodes arranged in a distributed architecture greatly improves network management capabilities. In this way, controller clusters are able to support mass device management and transaction consistency on large-scale hybrid cloud networks.
• Controller federation: In cross-DC, cross-region hybrid cloud scenarios, a single controller cluster managing multiple DCs will result in problems such as limited scale and poor reliability, which can easily result in ‘split-brain’ issues. The SDN controller federation architecture can completely solve this problem, providing support for cross-cloud application data migration and cloud service expansion. The controller federation architecture provides both scalability and processing performance. Domain controllers are responsible for local control-plane processing, specifically providing low-latency assurance. A ‘super controller’ is responsible for the processing coordination of multiple domain controllers in order to implement network expansion. 

Figure 1: Federated clusters for network flexibility

E2E Control in any Scenario

Hybrid cloud networks consist of cloud access networks and interconnected public and private cloud networks that span multiple networks, including campus, WAN (IP/MPLS and optical transport), and DC networks. Based on domain-specific control, the SDN solution can implement End-to-End (E2E) control in any scenario and provides network-wide resources on demand, automated deployment, and intelligent traffic control. To support more complex intelligent applications, coordination between domain controllers through orchestrators or super controllers is required. Due to the differing characteristics of each application scenario, the SDN controller architecture must have a flexible framework. 

Figure 2: On-demand deployment and intelligent traffic control

SDN Northbound Openness

Northbound openness is one of the most important features of the SDN architecture. Northbound openness gives SDN networks incomparable advantages over traditional networks, including:

• Network programmability: This transforms network resources and capabilities, making services open to users, and so accelerating service innovation.
• System integration: The controller can be conveniently integrated with the orchestrator, cloud management platform, and BSS/OSS through open northbound interfaces.
• Evolvable platform: Apps can be decoupled from the controller platform, achieving backward compatibility, without platform replacement or large-scale reconstruction.

To achieve northbound openness, the controller needs to provide not only network models based on defined open standards but also those based on the de facto standards of open-source platforms. 

Southbound Multi-vendor Heterogeneity

To achieve network programmability, the SDN controller uses southbound network control technology to manage and control the entire network device layer, including topology discovery, status collection, policy management, network configuration, path optimization, and forwarding entry delivery. The network structure is complicated and involves a wide variety of equipment vendors, and the goal of carriers and enterprises is to avoid vendor lockout. SDN controller southbound interfaces must:

• Support multi-vendor heterogeneity
• Mask the differences between interfaces on devices from different vendors
• Abstract physical devices to facilitate the management of pooled logical resources
• Support interconnection with third-party VAS devices to provide value-added services

To support multi-vendor heterogeneity, SDN controllers need to provide a southbound-driven mechanism and adopt plug-in technology so third-party vendors can flexibly develop customized plug-ins that support dynamic loading. 

High Network Reliability

SDN networks are a restructuring of traditional networks and involve a moving away from distributed control to centralized control. However, one of the downsides to centralized control is that controllers as the centralized control points may experience faults, and links between controllers and their respective managed networks may also encounter faults. As networks are the basic infrastructure of telecommunication systems, network disruptions and forwarding failures are of zero tolerance.

Solving the issue of SDN reliability is a progressive process involving the following steps. First, controller products must come with their own multi-layered, reliability-ensuring systems. This should include the use of distributed systems to achieve cluster reliability and geographical redundancy, the use of load balancing to increase northbound reliability, and the use of southbound transaction and reconciliation systems to achieve data consistency between the controller and forwarding plane.

Figure 3: SDN architecture with Nginx server

Second, the reliability of the connection channel between the controller and forwarding plane must also be ensured. For this purpose, redundant links are often used in order to ensure uninterrupted access between the network and controller. 

SDN Network Intelligentization

Network intelligentization is the key difference between SDN networks and traditional networks. As a network’s ‘brain,’ SDN controllers are the key components in the process of network intelligentization and provide the following capabilities: E2E agile interconnection from a global network perspective and intelligent on-demand bandwidth optimization, and intelligent network O&M with rapid troubleshooting and self-healing capabilities as a result of network resource pooling.

The SDN controller provides network optimization solutions that are flexible, efficient, intelligent, and in real time. By having a global network view, it can provide E2E application-based optimization for enterprise branches, WANs, and DCN networks. To perform intelligent network control, the SDN controller must be able to support network-wide topology discovery and service status (bandwidth, latency, jitter, packet loss rate, etc.) monitoring. Based on real-time network and service data, and user-defined network service requirements, the SDN controller uses global path optimization algorithms to implement network traffic scheduling and achieve network service perception and Bandwidth on Demand (BOD).

Figure 4: SDN O&M intelligence functions

SDN approaches the task of intelligent network O&M from a usability perspective, utilizing a diverse set of fault-detection methods, Big Data analysis platforms, and intelligent fault-decision-tree analysis to provide an efficient, intelligent, rapid, and self-healing O&M framework. This frees O&M personnel from their previously tedious workload and creates a framework for the automation of O&M-related tasks.

SDN Solution Deployment Phases and Commercial Practice

Total network transformation cannot happen overnight, and SDN deployment plans must adopt a flexible evolution pathway. This must take place while maintaining a focus on clouds as they extend from individual network scenarios to cross-network scenarios and then to E2E network scenarios, eventually achieving E2E NaaS. Deployment should be divided into three phases — ‘placing points,’ ‘connecting points to lines,’ and ‘forming a plane from lines’ — in order to facilitate hybrid cloud construction with a network-wide SDN solution. 

Phase I: Placing Points — Deployment of Domain-specific SDN Solutions

Technological development always requires you to start with the basics and gradually increase complexity over time. Any carrier, enterprise, or equipment supplier wishing to take part in SDN development must begin from a single domain, which means a single DC, WAN, or campus network. Dedicated enterprise lines to cloud servers also begin with overlay solutions directly creating site-to-site and site-to-cloud overlay tunnels in order to open dedicated lines rapidly and avoid the complexities associated with cross-domain WAN and cross-vendor and cross-technology interconnections.

Phase II: Connecting Points to Lines — Cross-Scenario Unity, Point Infiltration, and Application Expansion

After the successful implementation of the domain-specific solutions, carriers will gradually stitch together the deployment of cross-domain networks, extending the application range of SDN. This includes the following combinations:

• DCN and DCI orchestration, achieved by integrating the DCN and WAN to implement cloud-network synergy, which is the dynamic scheduling of network resources based on cloud application requirements (bandwidth, latency, etc.)
• Multi-domain control, achieved by coordinating the backbone, metro, and campus networks into one network to implement E2E automation and cross-domain traffic optimization
• IP + optical synergy, achieved by managing optical, electrical, Ethernet, and IP resources as a unified resource pool to improve network resource utilization 

Phase III: Forming a Plane from Lines — Implementing E2E NaaS with Cloud-Pipe-Device Synergy

With cloud-pipe-device synergy, the evolution of the bearer network from overlay to underlay, and the management of the entire network as a resource pool, network on demand for cloud connect will finally be achieved. Networks will be able to adapt dynamically to the demand of cloud services, providing E2E QoS and SLA, and providing high-quality connections for 4K video, video conferencing, the IoT, 5G, and financial services.

SDN has already been proposed for several years, during which Huawei has done a lot of exploration work, especially in cloud service scenarios where many commercial practices have been completed. Such cases include Deutsche Telekom (DT) and China Telecom.

In March 2016, DT and Huawei jointly announced the official unveiling of an open telecom cloud. DT chose Huawei to provide hardware and software solutions for its telecom cloud. When customers migrate their IT services to the cloud, they can perfectly balance prices, services, and quality. The SDN-based hybrid cloud deployment solution will further help customers carry out their hybrid cloud strategy.

In 2016, Huawei assisted China Telecom in constructing the IDC backbone bearer network by providing an SDN DCI solution that matched the innovative E-Surfing cloud-network synergy mode. The Huawei SDN DCI solution provided a high-performance dedicated cloud bearer network with ultra-wide pipes, ultra-low latency, network on demand for cloud connect, and cloud-based network resource sharing.

The development of cloud services is a long-term process, and the hybrid cloud will be the fundamental and long-term form of the enterprise IT infrastructure. The hybrid cloud requires ‘on-demand networks across private and public clouds’ to achieve agility, and SDN networks are undoubtedly the best choice. Huawei’s leading, open, scalable, and highly reliable SDN controller and network equipment solutions meet the development needs of the hybrid cloud, enabling customers to build an agile IT system to advance the development of their business and enjoy the true value of clouds.