New Thoughts on network architecture of data centers

The achievements of modern science and technology have greatly extended the space of human organs. If we say that telephone is an extension of human voice, TV is an extension of human vision, and data center is an extension of human brain, then, the network is a nervous system that exchanges information and sends instructions between various extensions and between the extension and the subject.

A normal nervous system is one of the basic symbols of a healthy human being. Similarly, A high-bandwidth, low-latency, and high-reliability network system is also a basic guarantee for the effective extension of various human organs and the entry into the cloud era.

I. Challenges of cloud services to networks

As information services enter the cloud era, more and more users enter the network through different types of service devices, and the amount of information exchanged between various service devices and between many users is growing rapidly, the types, quantities, and physical identifiers of various access devices are also growing rapidly, how to quickly and accurately transmit these large quantities of different types of information between various types of service devices has become a huge challenge for the network.

2. Overlapping Service Processing Methods in traditional network architecture

As a terminal device that provides services directly, the designer develops and uses the corresponding physical interfaces and communication protocols based on their business characteristics. For example, GE/10GE, FC, IB and other physical interfaces and communication protocols such as Ethernet, IPv4, IPv6, FCoE, and IBoE. Each protocol has its own specific data format, and corresponding control packets. In traditional network architecture, in order to connect terminal devices of different service services, various devices in the network are forced to support these interfaces and protocols simultaneously from hardware and software. At the data layer of the network device, you must identify the type of the received data, and find the corresponding forwarding rules based on the results for forwarding. In terms of Device Control, protocols for each type of services must be processed separately to coordinate and control forwarding behavior at the data layer. In this way, multiple business logic networks are formed in a physical network. As a result, as the number of service types increases, the business logic network is becoming more and more complex and software and hardware become increasingly complex. As a result, the network forwarding efficiency decreases and the development time of new businesses increases. In a data center, not only must the business logic networks of tenants be supported, but also the business logic networks of multiple independent departments may exist in a single tenant, as a result, the physical network of a data center needs to support a large number of business logic networks, making traditional overlapping services difficult.

Figure 1. Overlapping service model in traditional networks

Iii. New Ideas on Network Architecture in data centers

How can a physical network with a large number of nodes support a large number of business logic networks, and each business logic network may have its own address encoding system and transmission rules? Many examples in our lives have already given us the answers we need.

The postal code and standard email encapsulation are widely used in the current postal system. The postal code is the address code system in the postal system and the address of each service point in the postal system. The zip code on each email indicates the address of the target service point. In the postal system, you only need to identify the target postal code. Therefore, the Postal Code encapsulates the actual physical address of the user, decouples the user address from the transfer address, and solves the problem of rapid growth in the number of physical addresses of the user. At the same time, standardized encapsulation of nonstandard postal items also makes it possible to use machines for automatic and high-speed sorting during mail delivery.

In the cargo transportation system, the standard container size and the target address greatly simplify the intermediate links in the transportation process and improve work efficiency.

Like the postal and cargo transportation systems, the network is used to transfer user information from the source address to the target address. The address system and format of information vary with services, but the information transmission function of the network will not change. An efficient network information transmission system will not and should not be affected by the system and format of the transmitted information address. The network information transmission system only standardizes the transmitted information on its edge service nodes, and the transmission process within the system is not affected by the transmitted information content. From the perspective of the overall network architecture, the network information transmission system is like a four-in-one bridge, connecting the edge service nodes of the network, 2. The source node of the edge service only needs to add the address of the target edge service node according to the requirements of various services, and then deliver the transportation tools running on the bridge. On the bridge, only one type of efficient transportation means can be run to load standard containers.

Figure 2. Forwarding and business separation model in the bridge network

The bridge network separates service functions from the forwarding function, which not only solves the impact of rapidly growing user addresses on network routing, but also greatly simplifies the forwarding process between intermediate nodes in the network, improves network efficiency.

Iv. Basic working principles of the bridge switching network architecture

1. The business access nodes at the edge of the network encapsulate different types of data packets from the user in a unified manner, add the target and Source Business node information, and then send it to the nodes in the bridge network.

2. the nodes in the bridge network directly forward data based on the addresses of the target business nodes, decoupling the addresses used in the forwarding process from a large number of actual user addresses;

3. The target edge service node unpacks user data packets in a unified data format and sends the original data packets to the target users;

Compared with traditional overlapping service methods, the bridge switching network architecture uses the routing protocol computing results as the basis for internal data forwarding in the exchange network, and the Business Node Address at the edge of the network serves as the data forwarding address, the data forwarding address is decoupled from the user's actual address, eliminating the impact of user address changes on internal network forwarding. At the same time, only one standard format is used for data transmission between bridge nodes, which facilitates high-speed and effective data transmission, implements lossless Ethernet protocols and functions such as DCB, and facilitates simplified network management.

From the service perspective, the bridge network provides point-to-point, point-to-point, and multi-point-to-point connection services. The entire network is like a switch, router, FC switch, or IB switch ,..., The service node port on the edge of the network is the user port of the vswitch. For all the data, you only need to configure the correct outbound port address, the data can be delivered through the bridge network, without the need to care about how the specific implementation is within the network.

5. Selection of technologies used for bridging Networks

The data center network already has a variety of solutions and protocols that separate network services from data forwarding, such as GRE, NvGRE, VPLS, VxLAN, and MACinIP, and TRILL, SPB and so on. In addition to the basic requirements of the network, we hope that the selected technologies and standards can:

1. simple, convenient, and dynamic support for point-to-point, point-to-point, and multi-point-to-point connection services;

2. Supports end-to-end shortest path;

3. As many Logical Networks and multicast and broadcast groups as possible to meet the needs of many business Logical Networks;

4. compatible with existing data formats of multiple protocols to facilitate seamless business communication;

5. Not only can CT solutions be provided directly, but also various IT solutions should be effectively supported;

Generally, the data layer in the network determines what kind of work the network can accomplish under this Protocol, while the control layer enables the work to be completed automatically and effectively. The data layer plays a decisive role when appropriate standard protocols are selected.

From the data layer of the protocols used in the current data center, the IT solution is based on the existing two/three-layer network (such as Ethernet + IP Network), for example: GRE, NvGRE, VPLS, VxLAN, and so on. Their solutions are also designed to solve the defects and deficiencies of the existing network and a specific problem. However, the essential problems of the existing L2/L3 networks have not been effectively solved, and the potential resources of the networks have not been fully utilized.

The CT solution represented by TRILL and SPB tries to solve the basic problems of L2 Networks in essence. In actual implementation, the CT solution needs to update network devices, which is very different from the rapidly growing market demand of data centers. Therefore, while solving the basic problems of L2 Networks, the CT solution must consider the problems of L3 networks and how to effectively support existing IT solutions and work with them.

From the perspective of CT equipment suppliers, the basic problems of the two/three-tier network not only exist in the data center network, it is also widely used in enterprise Campus Networks, operator LAN, wireless access networks, and other network systems. It is of great significance to fundamentally solve the basic problems of Layer 2/Layer 3 networks.

6. Bridge architecture to build an "elastic cloud network"

The traditional network architecture, edge and core have similar functions. In this network architecture, if the cloud services, servers, and storage carried by the Network change frequently, You need to upgrade the entire network, in order to support new features, it will cause high network maintenance complexity. In addition, the network performance will be greatly affected when many complex services are stacked on the device.

Drawing on the successful experiences of Internet architecture, Huawei proposed the concept of "elastic cloud network" and adopted an advanced bridge switching architecture to deploy complex functions on the edge of the network to keep the core network concise; in addition, the business control layer is added to remove the easy-to-change business from network devices.

The "elastic cloud network" combines the high performance and complex business processing of the network. When cloud computing brings various new businesses, the entire network device does not need to be changed, but only needs to upgrade the business control layer, you can deploy new data center services with a stable network architecture for a long time. For example: only need to update IPv6 control plane can be smoothly migrated to IPv6, to ensure the stability of the core layer; only need to upgrade the FC network control plane can achieve the FCoE protocol upgrade, from FC-BB-5 to FC-BB-6, you do not need to change the network core layer.

VII. Impact of future new technologies on Bridging networks

Network switching is evolving from optical transmission and electrical switching to optical transmission and optical switching. Due to technical restrictions, it is technically difficult to directly modify the optical data in the optical signal within a short period of time, therefore, Direct Optical Signal Forwarding (end-to-end Routing) in optical switches is still the most simple and practical technical solution, the end-to-end forwarding mechanism in the bridge network architecture can well adapt to the development needs of this future technology.