When the network architecture of the data center is restructured

When the network architecture of the data center is restructured

The network is the "highway" of the amount of data in the data center, and the construction of these "highways" must be planned and designed in a unified manner to give full play to the advantages of network interconnection. A network with advanced architecture can save costs, avoid frequent failures, and facilitate O & M, which is crucial to the development of data centers. After decades of development, the network architecture is constantly evolving to adapt to the business deployment and development of data centers. However, before the emergence of new technologies such as cloud computing, big data, and virtualization, the network architecture has not changed much. In the past, the classic layer-2 network, VRRP + OSPF + STP protocol combined into a common architecture for each data center. It has been inherited by network engineers for more than 40 years and is easy to use, become a classic network architecture. All of this will be broken with the advent of new technologies. The most closed network in the data center must be changed to adapt to the development of new technologies. A new chapter in the network architecture has arrived, the network architecture has undergone innovative changes in five aspects.

First, the evolution from a classic network to a VPC

In a classic network, all users share the public network resource pool. No logical isolation is performed between users. The intranet IP addresses of users are uniformly allocated by the system, and the same Intranet IP addresses cannot be allocated to different users. VPC (Virtual Private Cloud, Private Network) provides users with a logically isolated Virtual network space. Within the VPC, users can freely define network segment division, IP addresses, and routing policies, you can create a subnet in the VPC. You can add multiple VM instances to each subnet. Security provides network ACL and Security Group access control. Before the VPC appears, cloud users lack network management capabilities, which are not available in the classic network architecture. Therefore, VPC has higher flexibility and security, which is a recommended method by Major Cloud vendors.

Second, the evolution of stacked devices to independent device Networks

Stack enables Virtualization of network devices. Multiple devices are virtualized into one device, which facilitates network management and simplifies the network structure. It is an innovative technology that has been widely used in the past decade. However, the stack technology is a private technology. devices of various network vendors cannot be stacked, which limits the evolution of the stack technology. At the same time, although the stack device improves the network backup capability, if a problem occurs, it can also be caused by the stacked device, but it is not as secure as an independent device, in particular, network services must be interrupted when software upgrade is required for Stack devices. This is unacceptable for data centers that provide cloud services. Therefore, the stacking technology is increasingly not suitable for data centers with high reliability requirements, and the data center is de-stacked. Of course, now we cannot go back to the old path ten years ago. We use classic OSPF/BGP and other routing protocols to achieve redundant network backup. The switching speed of these protocols is hard to satisfy, as a result, 1BR, M-LAG and other loosely coupled technology, network equipment is independent, and through these loosely coupled technology and these independent devices together, to achieve smooth network switching, the network architecture is more elastic, which reduces the fault caused by stacking and provides higher reliability.

Third, the basic network changes from layer-2 forwarding to layer-3 forwarding.

In L2 Networks, broadcast storms or loops often occur. Using a layer-3 approach can not only reduce loops, but also improve network bandwidth utilization. The second layer of a classic network uses STP, RRPP, PVST, and other ring network protocols to block many network links. The network bandwidth needs to be halved, resulting in a great waste of network resources, with the continuous expansion of the network scale, the convergence speed of these ring network protocols is not satisfactory, but there is no such problem in the layer-3 network. Although the three-tier network configuration is complex, it can be deployed automatically through the Controller. As long as the port of the network device is connected through a cable, all network configurations are automatically deployed. For example, in a VXLAN network, as long as the network device is configured with a route to connect to the Controller, the interface address is connected, and the Controller automatically sends a VXLAN configuration when it perceives a port UP with a connection server, the network is automatically interconnected, and the layer-3 configurations are complex and do not involve human intervention. Related configurations are already available in the pre-configuration of the controller and are automatically delivered.

Fourth, the network bandwidth is getting higher and higher

The speed of a single port changes from 40G/200G to G/G. Currently, 40G internal interconnection in the data center is very popular, almost becoming the standard for cloud data centers, and there are also many network outlets using G, this situation is rare three years ago. We can see how fast the network bandwidth is developing, and this situation will change soon. Broadcom recently released Tomahawk 3, the industry's first 400G-based chip. This single chip supports 32-port 128G, 64-port G, or-port G forwarding, increase the single port speed to 400 GB. You need to know that Broadcom has always occupied more than 90% of the market share in the commercial chip market of Ethernet switches. Some time ago, it also needed to acquire the chip giant Qualcomm, with the boost capability of Broadcom, I believe that a network device with a single port of GB will soon appear in the data center. Shannon theory confirms that the greater the bandwidth, the greater the signal-to-noise ratio. When the port speed reaches 1000 GB, it cannot be used. Therefore, the bandwidth limit of a single port is GB. There is still a distance from GB, and the network bandwidth of a single port is still increased, in the future, we will continue to improve.

Fifth, reduce the complexity of the basic network

In the past, spider-type network interconnection machine rooms no longer exist, and network devices are simply connected to each other to form a tree-like network structure. The configuration of network devices is simplified and unified, and the keyboard experts who operate on network devices at high speed are no longer needed. There is no need for complex connections between network devices. Everything is done by the various configurations issued by the Controller. In this case, the network relies on SDN and NFV. SDN is an innovation oriented to the network architecture, and NFV is an innovation oriented to the device form, using these two tools makes it extremely simple to configure and deploy network devices, or directly distribute the forwarding table through the Controller, or directly issue the VXLAN configuration through the controller without manual intervention. People only need to click on the Controller to complete business deployment, daily monitoring and troubleshooting.

Cloud computing brings vitality to the network architecture. network restructuring has become a hot topic in the development of data center technologies. The data center has ushered in the era of network reform. With the evolution of the network architecture, the network has changed from closed to open, and is more suitable for the future development of data center services.