SDN FOR CLOUD COMPUTING

In a cloud DC, virtualized resources are further abstracted as services for flexible use. Cloud computing services can be classified into Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS), which correspond to hardware, software platform, and application resources, respectively.

The requirements for quality attributes vary according to the cloud service layer. IaaS is dedicated to providing high-quality hardware services, while SaaS and PaaS emphasize software flexibility and overall availability. Based on the hierarchical decoupling and mutual distrust principles, they decrease the reliability requirement for a single service to 99.9%, meaning a service can only be interrupted for less than 8.8 hours over the course of a year. For example, users may encounter one or two malfunctions in the email system, instant messaging (IM) software, or even the OS, but no real faults in hardware systems or driver software.

In terms of software technologies, the software architecture and technology selection of cloud services offer varying quality attributes. Legacy software can be classified into IT and embedded software. IT software is applicable to the SaaS and PaaS layers, and focuses on elastic expansion and fast rollout. In fault recovery scenarios, or those that require high reliability, IT software uses methods such as overall rollbacks and restarts. Embedded software focuses more on the control of software and hardware statuses to achieve higher reliability, and is more widely applicable to the IaaS layer.

As IaaS systems become more automated and elastic, and as new software technologies such as distribution, service-orientation, Cloud Native, and Service Less continue to emerge, SaaS/PaaS systems are being subverted, and IT software is undergoing an accelerated transformation to Internet software. Based on the DevOps agile development mode, as well as technical methods such as stateless services and distributed computing, SaaS/PaaS systems provide self-service, real-time online, and quick rollout capabilities for services. This transformation drives the development of new Internet business models.

At the same time, users are beginning to re-examine whether IaaS systems have high requirements on real-time performance across all scenarios and whether refined control over systems is required. This kind of thinking also influences the development of SDN, splitting system development along two different paths: control-oriented and orchestration-oriented.

• 1. In the current phase, upper-layer services cannot be completely stateless, and the network still needs to detect the computing migration status. In addition, a unified management platform is required to implement fine-grained status management for routing protocols and other information, in order to meet the requirements for fast network switchover in fault scenarios and ensure no impact on upper-layer services. During the enterprise cloudification process, the following three solutions are provided for the control-oriented path, each of which can be chosen depending on software capabilities and the organizational structure of individual enterprises.
• • Cloud-network integration solution: This solution utilizes the open-source OpenStack cloud platform and commercial network controllers to centrally manage network, computing, and storage resources, and implement resource pooling. The cloud platform delivers network control instructions to the network controller through the RESTful interface, and the network controller deploys the network as instructed.
• • End-to-end cloud-network integration solution: This solution is also based on a combination of the cloud platform and network controller, and is typically used by commercial cloud platforms such as VMware and Azure. Compared with the open-source cloud platform, the commercial cloud platform offers improved availability and maintainability. As a result, most enterprises use this solution, with only those possessing strong technical capabilities preferring the open-source cloud platform.
• • Virtualization solution based on a combination of computing virtualization and network controller: This solution is based on a combination of the virtualization platform (such as vCenter and System Center) and network controller, with no cloud platform used to centrally manage computing and network resources. After delivering a computing service, the virtualization platform notifies the network controller, which then delivers the corresponding network service.
• 2. The orchestration-oriented path depends on future upper-layer service software architecture, and the expectation that IaaS software will be further simplified after SaaS/PaaS software becomes stateless. The status of the IaaS software does not need to be controlled. Instead, IaaS software only needs to be orchestrated using software tools, similar to the upper-layer system.

Selecting the appropriate path or solution depends on many factors such as the enterprise organization structure, existing software architecture, technology and resource investment, and cloudifica- tion progress. Although all four solutions have their strengths and weaknesses, they can support enterprise service cloudification over a long term. The following describes what features each solution offers and why an enterprise might select them.

• 1. Cloud-network integration solution: (OpenStack+network controller): This solution has the following strengths:
  • - As OpenStack is the mainstream cloud platform, the open- source community is active and provides frequent updates, enabling rapid construction of enterprise cloud capabilities.
  • - Customized development can be applied to meet enterprise business requirements. These are subject to independent intellectual property rights.
  • - OpenStack boasts a healthy ecosystem. Models and interfaces are highly standardized, and layered decoupling facilitates multi-vendor interoperability.

This solution has the following weaknesses:

• - Commercialization of open-source software requires software hardening and customization. As such, enterprises must already possess the required technical reserves and continuously invest in software development.
• - During the current evolution from single to multiple DCs, and to hybrid clouds, a mature multi-cloud orchestrator does not yet exist within the open-source community. This will need to be built by enterprises themselves.
• - Enterprises that choose this solution generally possess advanced software development and integration capabilities, are concerned about differentiated capabilities and independent intellectual property rights, and require standardization and multi-vendor interconnection. Currently, this solution is mainly used for carriers’ telco clouds, and for the DCs of large financial institutions and Internet enterprises.

2. End-to-end cloud-network integration solution (Huawei FusionCloud, VMware vCloud, and Microsoft Azure Stack):

The advantages of this solution are its full support for single/ multiple DCs, hybrid clouds, and SaaS/PaaS/IaaS end-to-end delivery, which enable the rapid construction of cloudification capabilities for enterprises.

However, areas where this solution requires improvement include openness and vendor lock-in.

Enterprises that choose this solution are often in urgent need of cloud services. They need to quickly build these services in a short period of time to support new business models and quickly occupy the market, but lack the required technical reserves. As such, this solution is typically employed for DCs of small- and medium-sized enterprises.

3. Virtualization solution (computing virtualization+network controller):

This solution has the following strengths:

• - In the virtualization solution, computing and network resources are independent of one another, and the enterprise organization (IT and network teams) does not require immediate restructuring.
• - A large number of cloud and PaaS/SaaS platforms are currently available, and the industry is considered to be mature. Consequently, the association between computing virtualization and network controllers enables the rapid construction of IaaS platforms capable of implementing automation and satisfying service requirements. This approach is low risk and is easy to initiate.
• - The open architecture of the IaaS layer and layered decoupling allow flexible selection of the cloud and PaaS/SaaS platforms.
• - This solution is based on mature commercial software, which requires no customized development and offers high reliability.

A weakness of this solution is that IaaS/PaaS/SaaS software models need to be selected in rounds, which slows down the cloudification of enterprises.

Enterprises that choose this solution have complex organizational structures and fixed service applications. They do not want to lock vendors, but have major concerns relating to solution stability and reliability. As such, this solution is typically implemented in the DCs of medium- and large-sized enterprises in the transportation and energy industries.

Following the development of container technologies, compute and storage resources are becoming less dependent on networks. In this solution, the network controller can evolve into an orchestrator, and the overall solution can evolve to tool-based orchestration.

4. Tool-based orchestration solution:

This solution has the following strengths:

• - Provides customized service orchestration capabilities to adapt to enterprise service applications.
• - Easy to develop and quick to launch, as it is based on script or graphical orchestration tools.

This solution has the following dependencies or weaknesses:

• - The IaaS, PaaS, and SaaS software must comply with the stateless principle, and service reliability is independent of regions and does not rely on the IaaS layer (VM migration).
• - Service application scenarios are relatively simple and services are independent of each other, preventing conflicts and mutual coverage impacts.

Enterprises that choose this solution do not rely heavily on legacy service software, or can begin restructuring at low costs. They operate clear service scenarios, and require rapid responses to service changes. In addition, they can apply strict DC construction specifications to ensure service independence. This solution is typically implemented in the DCs of Internet enterprises.

Enterprises have varying concerns about their network capabilities when choosing from the available solutions, as shown in

Table 1.2.

To summarize, enterprises should choose a cloud-based transformation solution capable of matching their specific service requirements and technical conditions.