Challenges facing data center network technology

The network is the most stable part of the data center, and the technology has not changed fundamentally for decades. However, with the popularization of the Internet and the rise of new technologies such as cloud computing and big data, the network has become the biggest stumbling block to the development of data centers. It is imperative to transform the network, but the network has a lot of historical baggage. The old Ethernet protocol is still the only widely used protocol in the current network (except for storage networks). To adapt to the development needs of the new situation and new environment, the network must introduce some new equipment and new technologies. This process is not smooth sailing, and network technology faces great challenges.

This article will discuss in detail the major technical challenges currently facing us. If they can be overcome as soon as possible, the future of data centers will definitely be bright.

[[216429]]

Connector SFI Technology

Data centers have higher and higher requirements for network bandwidth. It took almost 20 years to develop from 1M to 1G, less than 10 years to develop from 1G to 10G, and only five years to develop from 10G to 40G/100G. The bandwidth is being updated faster and faster. Single-port 200G equipment will also appear in the data center market in 2018. Putting aside the optical technology to achieve such high bandwidth, there are also many technical challenges to meet the line speed forwarding of single-port 200G equipment.

In fact, the most widely used device in the data center network is the switch. The switch that can provide 200G must be some large frame-type equipment. In the past few years, it was a common phenomenon that the network equipment could not reach the line speed inside. Therefore, it was often the case that some ports were not fully utilized at the line speed, but the equipment was already congested. That is, the internal equipment was not 100% line speed, which brought great inconvenience to the expansion of the data center. Therefore, the data center requires network equipment to be able to reach the line speed. If it is not possible to reach the line speed, it is meaningless to provide so many ports on a board. Therefore, the real line speed inside the network equipment is the basic requirement of the data center for the network. This poses a great challenge to frame-type network equipment, which must ensure that all board ports are forwarded at the line speed inside the equipment.

If a 48-port 100G card needs to be line-speed, it is necessary to provide 4.8T bandwidth for one card. If it is a frame-type device with 16 slots, then 76.8T internal bandwidth must be provided. This requires the deployment of a large number of connectors between the cards, and these high-speed connectors can achieve line-speed forwarding of network traffic between the cards. Connectors generally use the Switching Matrix Interface (SFI) technology, which can provide faster speed, more bandwidth and higher stability for the backplane and interconnection. At present, a general connector has a bandwidth of 60G. For a 4.8T card, 80 connectors must be deployed, which is not only a large number, but also takes up too much internal space, making the device too bloated. It is very important to increase the connection rate of a single connector. Now there are also some 150G and even 240G connectors. The higher the speed, the worse the stability. Simply increasing the speed of the connector will cause pin oscillation or packet error, which makes the device unstable. Therefore, connector technology is still being improved. While increasing the connection bandwidth, stability must be increased. Otherwise, it will be impossible to realize 200G or even 400G on the switch equipment. The connector part is the primary problem that needs to be solved urgently.

SDN controller implementation technology

SDN is an important technology for automatic network deployment and management. If it can be widely used, it will greatly save the investment cost of data center operation and maintenance. All data centers now have to reserve some network technicians for daily network maintenance and troubleshooting. SDN network can free people from the complicated and tedious operation and maintenance work.

The vision of SDN technology is beautiful, but the reality is cruel. Most of the current SDN technology comes from traditional network equipment manufacturers. Some software startups also provide pure software SDN controller software, but network equipment vendors do not buy it and do not actively support these software, making SDN controller software unable to be applied in practice. Network equipment vendors are not idle, and have also launched their own SDN controller software, but they can only achieve interconnection and control with their own equipment. This has led to the current situation of SDN deployment in data centers. Each network equipment vendor's SDN controller is sold in conjunction with network equipment. The SDN network has become a closed network. Once a data center chooses the equipment of an equipment vendor, it can only choose this vendor for subsequent expansion. This will be a situation that all data centers do not want to see.

However, promoting the unified standard of SDN controllers is equivalent to asking network equipment vendors to revolutionize themselves, which will inevitably lead to the demise of many equipment vendors, so the resistance to implementation is very great. Of course, SDN technology itself does not have any problems, but it is mainly the resistance of the network environment that makes SDN technology still not widely used in data centers.

Virtualization Technology

Virtualization is widely used in data centers, including the network part. Many network devices support virtualization (one virtual to many, many virtual to one), but the virtualization of network devices has also encountered many problems in actual applications.

When a network device is virtualized into multiple devices and rented to different users, although they appear to be isolated, they are actually still working on the same device, or even the same board. When the board or device is restarted, all user services will be affected. There is also a security issue. Although virtualization is logically isolated, they are still physically together and data traffic may be interoperable. In particular, when a network device is attacked, all virtual device services may be affected.

When multiple network devices are virtualized into one device, it is indeed more convenient to manage. However, once a fault occurs, the multiple virtualized devices are too tightly coupled, and it is inevitable that innocent people will be affected. Also, when upgrading the equipment, because multiple devices are tightly coupled, it is impossible to upgrade each device one by one, which will interrupt the business and make network maintenance extremely difficult. If the core network equipment of the data center needs to be restarted, it must be a major network accident.

. Virtualization technology has become outdated in the current development of data centers. Now there are some loosely coupled virtualization technologies, which are lightweight virtualization between devices and within devices. This makes it very easy to unlock each other and has very little impact on each other. However, these technologies are still being improved, and whether they can take root in data centers remains to be seen. In any case, the current virtualization technology is no longer satisfactory, and there is an urgent need for the emergence of new network equipment virtualization technology.

The above are several challenges currently faced by network technology, which are still being improved, but I believe that in the future, these problems can be well solved, so as to promote the development of data centers.