Future SDN-based Data Center Network

Internet data centers are growing fast as the internet expands. As more processing is done at the cloud end, cloud data centers are also growing fast. Internet giants such as Google, Microsoft, and Tencent each have more than 100,000 physical servers in their cloud data centers. In the future, internet traffic will be located within the cloud data center.

Challenges Facing the Data Center Network

As data centers expand fast and cloud computing is deployed widely, new requirements are being put on the data center network in terms of network management, service support, and energy efficiency.

Centralized, efficient network management

A large-scale cloud computing data center generally comprises tens of thousands of physical servers and hundreds of thousands of virtual machines (VMs). Such big server clusters require thousands of physical network devices and tens of thousands of vSwitch devices for interconnection. A large-scale data center network requires centralized management for efficient maintenance, and it also requires fast troubleshooting to improve network usability.

Flexible, efficient networking

Redundant links, protection links, and other proper protection mechanisms should be considered in the design of a complex, large-scale, cloud computing data center network. These guarantee network flexibility and reliability. Virtual Router Redundancy Protocol (VRRP) and Double-Link Upward Connection and Spanning Tree Protocol (SPT), which are both widely used in data center networking, only protect part of the network. This makes the network underutilized and vulnerable.

VM deployment and migration

A cloud data center is deployed with a large number of VMs that need to be flexibly migrated according to service needs. The data center network therefore needs to be able to identify VMs and adopt appropriate network policy according to VM deployment and migration.

Service support for multiple virtual tenants

The cloud data center should provide users with virtual private cloud leasing.  Tenants can configure their own subnet, VM IP address and access control list (ACL), and they can also manage their own network resources. The data center network needs to support multiple virtual tenants and guarantee tenant isolation and security.

All-round IaaS

Cloud technology helps the data center virtualize computing and storage resources and therefore provides users with IaaS concerned with computing and storage resources. Network resources are yet to be provided virtually on demand; that is, all-round IaaS, including computing, storage and network resources, is still unavailable.

Current Technologies of the Cloud Computing Data Center Network

Some technical solutions have been proposed to address the challenges faced by the cloud computing center.

Trill and SPB

Transparent interconnection of lots of links (Trill) and shortest path bridging (SPB) were designed to meet the demands of multipath forwarding and flexible data center deployment. They solve the problems of decreased link utilization ratio and degraded network stability created by shortest path tree (SPT) deployment. They also solve other complicated problems caused by layer-3 routing in the data center.

Trill and SPB allow for loop-free link counting based on the IS-IS protocol. They use equal-cost multipath technology for multipath forwarding with balanced loads. This increases the network utilization ratio and reliability of the data center network and makes network deployment more flexible. However, neither Trill nor SPB solve problems related to centralized management, VM deployment and migration, and IaaS provisioning in a data center network. 

EVB

Edge virtual bridging (EVB), defined by IEEE802.1Qbg, includes the edge relay, virtual edge bridge, and virtual edge port aggregator. EVB identifies and bears multiple VMs. It uses the VDP protocol for dynamic VM creation and migration discovery, and it automatically configures relevant network parameters to meet the needs of dynamic VM migration. However, EVB also does not solve problems related to centralized management, multipath forwarding, virtual multitenant, and IaaS.

VXLAN

Virtual extensible LAN (VXLAN) addresses virtual multitenant and VM migration issues in the data center. VXLAN uses the L2 over L3 technology to add VXLAN and IP encapsulations to the original packet encapsulation. In this way, the original L2 packet can pass through the L3 network, the scope of the L2 network can be expanded, and the VM migration can occur flexibly across L3. VXLAN encapsulation greatly expands the tenant ID field and overcomes the 4K capacity limitation of the VLAN mode. However, VXLAN does not solve problems related to centralized network management, data center, and IaaS. 

NVGRE

Network virtualization using generic routing encapsulation (NVGRE) is also designed to address the issues of virtual multitenant and VM migration in the data center. NVGRE also uses the L2 over L3 technology; however, unlike VXLAN, it uses GRE encapsulation and supports multiple virtual tenants by matching the tenant ID with the GRE tunnel. Centralized network management and IaaS issues are not taken into consideration either. 

At present, the main data center networking technologies are only designed to tackle specific demands within a data center. They only solve some of the problems of a data center network.

Data Center Networking Solution based on SDN

A software-defined network (SDN) has forwarding and control, centralized control logic, network virtualization, and open network capabilities. This fits well with the requirements of a data center network in terms of centralized network management, flexible networking and multipath forwarding, VM deployment and intelligent migration, virtual multitenant, and IaaS. Cloud data center networks based on SDN is a future trend.

A data center network solution based on SDN comprises the SDN controller, VM manager, and DC manager. The SDN controller controls and manages network devices, including the vSwitch that resides in the server. The VM manager is responsible for VM creation, deployment, and migration. The DC manager is responsible for overall control and coordination of computing, storage and network resources, i.e. mainly the coordination between the VM manager and SDN controller.

Centralized and efficient network management and maintenance

SDN has separate forwarding and control as well as centralized control logic. The SDN controller stores static topology and dynamic forwarding information about the whole network so that the network can be efficiently managed and optimized. This, in turn, helps network troubleshooting. With static topology and dynamic forwarding information about the whole network, dedicated fault diagnosis tools can be developed to simulate the actual forwarding process. This is useful for locating and handling faults quickly.

Flexible networking and multipath forwarding

In a data center network based on SDN, the forwarding rules and actions are controlled and delivered by the SDN controller, which asserts control through the forwarding flow table and according to service needs. This prevents loops and ensures multipath forwarding and balanced load so that network reliability and usability are greatly improved.

Intelligent VM deployment and migration

VM deployment and migration require network coordination. In a data center network solution based on SDN, the DC manager, VM manager and SDN controller work together to support intelligent VM deployment and migration. When a VM needs to be migrated, the VM manager senses the need first and then sends out a network coordination request to the DC manager. The DC manager receives the request and then sends a coordination request to the SDN controller. The SDN controller issues a proper network strategy to the destination network device and abolishes the original network strategy of the network device where the VM used to reside. In this way, coordination between the VM and network is seamless, and the VM is migrated automatically and intelligently.

Support for a large number of virtual tenants

To bear a large number of tenants, ZTE’s data center network solution based on SDN identifies tenants using current MPLS labels. The 20-bit MPLS label supports an extremely large number of tenants. The SDN controller can separate tenants so that every tenant has a virtual network view, resource control, and guaranteed security.

IaaS with coordinated computing, storage and network resources

With the help of the SDN controller, network resources can be virtualized and provided on demand. The DC manager helps coordinate the SDN controller and VM manager so that coordination between computing, storage, and network resources is seamless. Users can receive IaaS on demand. 

ZTE’s solution based on SDN helps in the deployment and operation of a large data center. SDN technology is still developing, and there are still some issues to be addressed. It is critically important to achieve seamless interworking between the DC manager, SDN controller, and VM manager provided by different vendors. Evolving from an existing data center network to a data center network based on SDN is a challenge in terms of network deployment, and technological changes and SDN trends require continuous attention.

Conclusion

Many solutions have been proposed to address issues with data centers. Interfaces for coordination between the DC manager, SDN controller, and VM manager should be standardized, and network devices that support multiple forwarding modes should be deployed so that the existing forwarding mode and SDN-based forwarding mode can function. SDN technology and architecture need to be continually monitored and appraised, and international standards need to be developed to ensure the continuity in the SDN architecture and compatibility with different SDN versions.