SDN Solution for Data Center Networks
By Jason Ning, Data Center Network Switch and Solution Planning Manager, Huawei Enterprise Business Group
As cloud computing matures, more and more enterprises are moving services to cloud-computing platforms. What will next-generation data center networks look like?
SDN offers a new approach to resolving data center network issues. For example, the Open Networking Foundation (ONF) has defined the SDN OpenFlow communications protocol, which separates the forwarding plane from the control plane. IT vendors have proposed the concept of software overlay. Can SDN resolve core issues such as these for data center networks?
Core Data Center Network Issues
In recent years, data center networks have undergone the following major changes:
• Transition from 1 GE servers to 10 GE servers;
• Transition from high-density 10 GE convergence to high-density 40 GE or 100 GE convergence;
• Proposed use of optical switching backplanes;
• Emergence of vSwitch and disputes about the network border;
• Virtual machine migration and changing demands for rapid network policy migration;
• Virtualization using large Layer 2 networks and shifting demands for Transparent Interconnection of Lots of Links (TRILL), Shortest Path Bridging (SPB), and software overlay;
• Transition from Communications Technology (CT) solutions to Information Technology (IT) solutions;
• Transition from a traditional solution to an SDN solution.
These changes have been discussed in various forums and exchanges; however, many issues remain. These issues are broadly categorized as follows:
● Compartmentalization between the Network and the Cloud
There are two major aspects to this issue:
(1) Network design and deployment are not typically associated with services. Except for service security and reliability, data center networks are independently designed, procured, and deployed. However, computing, storage, and network resources must be integrated and orchestrated to instantly provide users with on-demand cloud computing services.
(2) Open Application Programming Interfaces (APIs) on networks are bottom-layer interfaces, which make it difficult for cloud-computing services to invoke these APIs to provide important functions.
● Key Barriers that Prevent Networks from Supporting Cloud-Computing Services
If experts agree that the essence of networks is connectivity, what connections are necessary for cloud services? What other issues and challenges must be dealt with before networks are capable of effectively supporting cloud services?
Here are a few important issues under discussion:
(1) Reducing network congestion. As the enterprises' own data volume and the horizontal traffic volume that comes with cloud computing increase, and servers' transition from 1 GE to 10 GE interfaces, data center networks must enable congestion-free switching to ensure consistent quality of services.
(2) Network virtualization. Cloud computing services necessitate the scheduling of network resources. Therefore, networks must support virtualization, especially for tenants' network resources, which include devices, Layer 2, Layer 3, and value-added services. Virtualization for resilient Layer 2 networks depends heavily on new devices and technologies, such as TRILL, SPB, Virtual Extensible LAN (VXLAN), and Network Virtualization using Generic Routing Encapsulation (NVGRE). These technologies, however, are immature and non-serialized due to their strong reliance on ASIC chips.
(3) Network experience. Traditional networks are characterized by best-effort and shared forwarding. Indeed, there is no guaranteed way to ensure end-to-end quality and effective fault location. In the cloud-computing era, enterprises are moving more services to cloud platforms, and device services are increasingly dependent on cloud center services. It is therefore essential to deploy networks with quality assurance and proactive fault location.
(4) Network evolution. A primary concern for enterprise customers is how best to migrate to SDN networks in a way that protects current network investments. Are there viable SDN solutions that interoperate with existing networks? Are current networks being built SDN-ready?
To better resolve data center network issues, Huawei has been actively exploring SDN technologies and engaging in joint SDN innovation with partners. Collaborations include Tencent SRP (Sequoia Routing Protocol) and Microsoft NVGRE. Based on the strengths and practices of SDN, Huawei approaches data center network issues with these design innovations:
● Cloud Service-Driven Hierarchy
Different user types and services present a range of concerns and requirements centered on network services, resources, devices, and forwarding stream customization. In a cloud service-driven hierarchy, the SDN architecture is divided into the management orchestration layer, controller layer, and device layer. Every network layer is abstracted, and network capabilities at each layer (including device, resource, and service layers) are open.
User-facing interfaces are simplified so that users can select APIs in different layers for different service demands. The Huawei Enterprise Software Development Kit (eSDK) contains these APIs and uses RESTful, OpenFlow, and NETCONF as key interfaces to rapidly adapt platforms to service needs. Interfaces at the resource or service layers can be used – depending on actual situations – for cloud services and platforms, enabling the rapid provision of cloud services. The use of interfaces at these layers also eliminates the need to directly invoke device interfaces, which is generally quite complicated.
Huawei also proposes using agile switches by opening the forwarding plane to:
• Provide greater network programmability;
• Reduce the amount of time it takes to update hardware (closer to software update time) and gear networks towards the fast-changing needs of cloud computing;
• Provide a hierarchical and adaptable SDN architecture capable of facilitating rapid service innovations for users.
● Unified Network Virtualization
Network virtualization aims to classify, schedule, and expand network resources in a flexible structure. In Huawei's SDN unified network virtualization, Layer 2/3 connections and value-added services are centrally integrated and managed by an SDN controller. At the network layer, unified overlay technologies are used to integrate intra-Data Center (intra-DC), inter-DC, and DC access and to manage physical and virtual networks from a central location. The SDN controller schedules network resources from end-to-end to facilitate fast provisioning of cloud-computing services. Unified network virtualization greatly simplifies network architectures and bearer technologies while ensuring flexible access at the edge layer and high-performance forwarding at the core layer. Moreover, unified technologies eliminate complex networking conversions and make networks simpler to manage, more efficient, and more easily scaled. With unified differentiation and orchestration under an SDN controller, device interconnectivity is further enhanced to guarantee fast connectivity and orchestration among multiple data centers.
● Cloud Service Experience Optimization
The Huawei cloud service experience optimization function monitors and optimizes the user experience in real time. It can be deployed in conjunction with cloud services and network resources. This function provides service customization for users and enables unified management and control in real time. To meet end-to-end quality and O&M requirements for cloud services, the optimization function instantly identifies service quality issues and dynamically changes network operations through central controls. This function implements all these features immediately: identification of service quality issues, fault location, network recovery, and guaranteed service continuity.
Huawei's proprietary Packet Conservation Algorithm for Internet (iPCA) presents a fresh approach to monitoring service quality and locating faults. When a network is being deployed, iPCA enables the network to monitor service quality through real-time monitoring of packet loss ratios, delays, and jitter on service paths. This function does not require manual intervention. It instantly reschedules network resources to ensure a consistent end-to-end service experience and dynamically adjusts the network as required.
Huawei's service-centric SDN architecture will be implemented in phases:
• Local monitoring and local recovery;
• Global monitoring and local recovery;
• Global monitoring and global recovery.
Huawei has resolved global monitoring issues using iPCA, which can be applied to any network or service and provides reference information and suggestions for local recovery.
● Sustainable, Smooth Evolution
Smooth evolution focuses on making full use of existing networks. Huawei uses a cloud controller (data center SDN controller V1.0) to provide a virtual machine migration automation solution for data centers. The cloud controller centrally manages and controls existing physical and virtual networks and connects to mainstream controllers over open APIs, implementing automatic network policy migration. Huawei will continue development of the cloud controller through cooperation with strategic partners to provide comprehensive SDN solutions that address user demands while fully utilizing existing networks. This solution will allow users to benefit from SDN earlier than expected.
Huawei will also launch agile switches with Protocol-Oblivious Forwarding (POF) to implement high-efficiency forwarding on SDN networks. POF frees forwarding devices from the need to support special protocols. By abstracting the controller and forwarding device interfaces to packet-forwarding instruction sets, forwarding devices are unaware of packet protocol types. All packet-forwarding functions are controlled by the controller's software. By specifying data offsets and lengths, the controller reads or writes packet data using generic instructions.
Decoupling software from hardware enables the forwarding and control planes to independently evolve, facilitating rapid service deployment without the need for hardware upgrades. In this way, Huawei's sustainable, smooth-evolution architecture actively protects user investments. Huawei provides mature SDN solutions to help users unleash the full potential of existing networks and benefit from the advantages that SDN offers. Huawei also offers an smooth-evolution SDN architecture that aligns customers' newly-built networks with future SDN developments through regular software upgrades.
Conventional wisdom indicates SDN will first be applied to cloud data center networks to handle key issues, such as compartmentalization between the network and the cloud, immature network virtualization, poor user experience, and barriers to network evolution. These issues prevent networks from supporting mature cloud services. An expert once commented: "Networks always lag behind."
Through joint innovation with partners, participation in standards organizations, and our own strengths in developing key technologies, Huawei has proposed the following SDN architecture for data center networks:
• Cloud service-driven hierarchy;
• Unified network virtualization;
• Cloud service experience optimization;
• Sustainable, smooth evolution.
SDN allows data center networks to keep pace with IT development regardless of performance or features by transitioning from closed to open networks, from hardware dependency to full programmability, from fragmented technologies to full quality assurance, and from complete overhaul to gradual evolution. Using our in-house programmable hardware platform and data center controller, Huawei SDN architecture implements key technologies such as iPCA and POF while making hybrid SDN available to existing networks. This ensures the system can perform instant monitoring, local traffic optimization, and rapid user service customization.
Huawei proposals include a gradual SDN evolution strategy that protects user investments by avoiding a complete network overhaul. With these innovative technologies, Huawei provides a stable, smooth-evolution SDN platform to usher users into the cloud-computing era.