Shulou(Shulou.com)06/01 Report--

The Development course of SDN Industry

Network technology is constantly improving with the rapid development of the computer industry, and information has evolved to the era of cloud computing, supporting the continuous emergence of hundreds of millions of user platforms. Human beings have set up super-large-scale data centers all over the world to meet this challenge. The network technology architecture layer, which was originally based on human control, is far from being able to meet this demand.

Macroscopically, the evolution of basic network architecture is slow.

At the micro level, hardware manufacturers are weak in innovation related to data forwarding and control.

Network hardware manufacturers are either content with the status quo or unable to initiate and promote innovation because of commercial interests. So SDN came into being.

These three characters can be regarded as the ancestor and founder of SDN. The one in the middle is Stanford professor Nick McKeown, the teacher of Martin Casado on the left, and Martin Casado is the founder of Open vSwitch. Scott Shenker is a Ph.D. in physics and a half-way monk in the field of the Internet. Among them, Professor Nick McKeown and Professor Scott Shenker put forward the interpretation of SDN from different angles, which is also the difference in the concept of ONRC and ONF between the two organizations. Professor Nick McKeown analyzes and solves the current network problems from the perspective of "system function reconstruction", while Professor Scott Shenker tries to solve the existing network problems from the perspective of "redefining abstraction". These two not only put forward the interpretation and concept of SDN, but also actively promote the commercial operation and industrialization of SND. Among them, these three figures collaborated in 2007 to establish Nicira, the first technology startup company focused on SDN, and began to try the commercial development of SDN. Professor Nick McKeown and Professor Scott Shenker also established the ONF organization in 2011 to actively promote the industrialization of SDN.

At present, from the perspective of industrialization development, the main players of SDN industry include, but are not limited to, the following:

The above figure lists the major players in the SDN industry, and after understanding this, we can discuss it separately, because SDN has developed around these characters over the years, including all kinds of love-hate entanglements between these characters.

2009-2011: the initial stage

We call it the beginning of the three years from 2009 to 2011, when Nick McKeown, a professor at Stanford University, first proposed the letter SDN at a conference, which is said to have been coined by a journalist. It used to be called OpenNetwork, but in 2009, Microsoft and Google were already doing SDN. When it is limited to internal research and use, unlike startups that treat SDN as an industry. Microsoft released a VL2 product that year, and Goolge released G-wan, which was officially launched in 2011, and uses G-wan technology to connect its 13 data centers around the world. At the same time, there have been many famous SDN startups in the past three years. As mentioned above, Nicira has tried to commercialize SDN for the first time. BigSwitch was founded by Stanford professor Nick McKeown and some colleagues in the same lab, and took a completely different path from Nicira. Nicira followed the path of pure software and was later acquired by Vmwar, a virtualization giant that also makes software, and is like-minded. BigSwitch went the other way: the white-card switch. In that era, in addition to the huge investment intention of these two companies, at least more than 20 SDN startups in Silicon Valley received investment. So it's not the age of a SDN tuyere.

At that time, the research team headed by Nick McKeown, a professor at Stanford University, initiated and promoted the famous network deployment model of separation of control and forwarding and the key protocol OpenFlow, which meant that the birth of Software defined Network (SDN) was officially announced, but the concept of SDN had not yet been put forward at that time. SDN in China also started in 2006, and there is a team studying OpenFlow synchronously in the Institute of Information Technology of Tsinghua University. Since 2009, the first domestic SDN system developed by Tsinghua University was officially launched. In 2011, the doctor of Tsinghua University, who led the development of this system, came out to establish the company Spruce Network. Together with Nicira and BigSwitch of that year, they got the investment of the same investment fund that year.

Because now everyone is watching SDN when looking at his local point in time, at this point to talk about the occurrence of technological phenomena or a specific industry. But in fact, when you understand the evolution of SDN and the game experienced by manufacturers in the middle, you will find what we should do now. What to do is an effective way, where to cooperate with manufacturers, where to cooperate with the community, and where to develop by yourself! When we see all these clearly, our SDN landing will be more smooth.

When it comes to networking, in the words of Silicon Valley, there are only two companies in the world that do networking, one is Cisco and the other is non-Cisco. In those years, Cisco's main product was the Nexus 7K, a data center product that also supported virtualization and virtual networks to a certain extent. At this stage, Cisco is still focused on buying hardware solutions, but the hardware is more in line with the trend of virtualization and data center construction.

At this time, Cisco's competitor Juniper spent $100 million on a partnership with IBM to launch the QFabric system shown in the figure below.

We can think of it this way, this is the first generation of SDN system launched by the manufacturer. Why? Because the concept of "controller" appears for the first time in the QFabric system, and the architecture is much simpler than Cisco's, using fewer connections to achieve more data center functions. There are many benefits of this architecture, as shown on the right side of the above figure, both financially and efficiently. With the use of the QFabric network, the data center network has become a large switch environment.

So this architecture itself has realized the flattening and consistency of the data center network. All nodes are on the edge, no matter your router, storage or server, all exist in the edge node, and the control of these networks is provided by the Fabric Director in the middle. So this whole set of architecture is close to the architecture design of a SDN system. Why don't you think of QFabric as the originator of SDN in the industry? It is because the system is a closed commercial system and is not open. When SDN was first launched, it came out of OpenNetwork, and he needed to unbind the vendor and use an open source system to do it. Due to these reasons, QFabric is not recognized by the industry, so it does not have a great impact on the industry.

The first one that has a great impact on the industry is Nicira. The following figure shows a network architecture diagram designed by Nicira. The ladder diagram above is the data center network, and below are virtual machines, physical machines, and so on. This is a common situation in data centers. If it is an access network, we can think that this is the entire campus network or ubiquitous network, and the following virtual machines are a community. In this case, the network design, if the virtual machine or cell network needs to carry out network isolation, security protection, traffic restrictions and other operations. It is necessary to configure and manage the rule entries independently of the network equipment, and when the entire configuration and maintenance engineer leaves after three years, the inheritance and maintainability of knowledge can not be well maintained. then the successor will artificially lead to some repetitive or new rules, then there will be more and more rules on the network equipment, making the network more and more complex.

This kind of human intervention leads to the network becoming more and more complex over time. In addition, more and more new network protocols and algorithms make the network control plane more and more complex. So that the network has developed for so many years, it is still in the stage of "management complexity", in order to make the current network have more programmable ability, so as to automate and intelligent network management. Nicira has developed a way to realize network management based on software so as to realize automatic and intelligent network management.

The following figure shows the architecture of NVP, Nicira's first SDN product:

The power of this product is so powerful that Nicira bought the price of US $1 billion without any revenue, which can be imagined in the degree of technological innovation and advantage of this technology and product in that year. In the figure, all virtual machines exchange and route network functions through Open vSwitch switches and controllers. OpenvSwitch is an open source virtual switch submitted by Martin Casado in August 2007. This open source virtual switch was really called Open vSWitch in May 2009 and officially entered the main stack of Linux in 2012. Due to the difficulty of unifying technical standards and conflicts of interest, other manufacturers launched component virtual switches, such as Cisco's Nexus 1000V vDS and IBM's DOVE were commercial virtual switch products at that time. Because the default Bridge in Linux can not be called virtual switch completely, it does not support OpenFlow and some advanced features of switch, which makes Open vSwitch get a high degree of attention, especially after merging into the main stack of Linux kernel, Open vSwitch has almost become the actual standard of open source virtual switch.

The NVP product realized the first network virtualization product independent of hardware and supported a variety of X86 virtualization architectures, leading or even subverting the network virtualization technology at that time. In NVP architecture, whether you are a LAN network or a data center network, as long as it is a three-tier IP. At this time, the Open vSwitch virtual switch running on the server virtualization or the physical switch supporting OpenFlow protocol is used as the subordinate connection point, and then through the rules built by the virtual network to achieve the function of interconnection between virtual machines. All the functions are implemented at the edge, which is in line with the original concept of network design, that is, all functions are implemented at the edge. In the past, the edge was the network equipment of Cisco and Juniper, but now the NVP architecture has smoothed this implementation into the server. The server provides more powerful computing power to do this edge computing. In this architecture, NVP does three main tasks:

1. The irrelevance of network equipment

2. Put Open vSwitch into the Linux kernel and use the computing resources of the server for network computing.

3. Realize the separation of control platform and data plane.

When the NVP architecture places Open vSwitch in the Linux kernel, the controller equivalent to NVP can control Linux systems all over the world. This kind of influence is one aspect of creating the value of NVP. In terms of engineering applications, NVP has achieved real practical applications, and ideologically, he has produced a revolution: decoupling software and hardware and making software and hardware independent.

In addition to the pioneer of startups, Internet companies like Google are also practicing and paying attention to the development of SDN. With the expansion of its network scale, Google is facing more and more challenges from network technology. Between the big data centers in Google, each data center has a link connected to each other, how to make the utilization of each line higher? In this way, the carrying capacity of the business will be greater, because each line is an undersea optical cable, and the cost is very high. So when Google was doing SDN, the utilization rate of each of his lines was less than 50%. This is very similar to when we design a network according to a certain capacity and retain redundancy, but after implementing SDN, the utilization of each line is less than 99%. The reason why Google dare to run 99% of the network in the SDN network is that they have achieved very fine-grained control over the network through SDN. The following is a picture of the 13 data centers identified by Google:

Google's SDN architecture is shown in the following figure:

The above diagrams explain the principle of Google's SDN architecture, G-WAN. In each of the thirteen data centers mentioned above, there will be dozens of switches that constitute sadly software-defined gateways. It is the network of such switches that allows Google to control 7% of the world's traffic. In the second picture, the architecture of G-WAN is revealed for us. The main idea is to achieve numerical control separation. The bottom controller is not affected by the upper controller. After the upper controller is disconnected or fails, the bottom can be cured completely. It is the healing principle designed by this controller that enables Google to use it to achieve global business.

At first, Microsoft focused on its own data center, and finally deployed it around the world. At the beginning, Microsoft also used a large number of white-card switches to build its own SDN network in its own data center. Unlike Google, Microsoft's main advantage is in the field of operating system and server, so Microsoft has done a lot of design and technical optimization on the network card. Finally, it leads to the difference between the two SDN. Goolge's SDN architecture is mainly implemented at the network level, while Microsoft's SDN architecture is mainly implemented at the system level. But the final effect is the same.

The following figure shows Microsoft's SDN architecture VL2:

The main contributions of VL2 here are:

1. The SDN architecture is mainly different from the traditional 7-layer network. It is a flattened network structure based on a two-layer structure. The two-layer structure expands infinitely and can be extended to three layers.

2. Only responsible for point-to-point accessibility, do not do anything else

3. All functions are realized by using the control of Flow at the edge.

At the same time, the first SDN practice in China was realized in Tsinghua University. At that time, the research team of Tsinghua University used the existing three-layer network of the Fit building, plus OpenFlow plus some TP-Link boxes, to refresh the firmware of the TP-Link box to enable TP-Link to support the OpenFlow protocol. At the same time, Open vSwitch is installed in the server, and when there is this structure, the whole building becomes a SDN network. All historical online activities can be monitored, you can let any traffic to do a firewall, when your firewall performance is not enough, you can also do multiple firewalls through the edge of the network load balance.

These are the first three years of SDN, and then, from 2012 to 2014, SDN entered a cold period. SDN startups were particularly confused during this period, and many SDN companies could not sell any products and technologies after they got the money to make good products. In retrospect, this period is a critical period when various standards, giants, Linux foundations and so on are playing games against SDN.

During this period, this fact was supported by several mergers and acquisitions, VMware's $1.2 billion acquisition of Nicria. Cisco's $1 billion acquisition of Insieme,IBM and SoftLayer,SoftLayer is designed to combine the strengths of Microsoft VL2 and Google's G-Wan. After acquiring Nicria, Vmware launched the whole number of NVP products and NSX products. Cisco has also launched the ACI architecture based on Insieme.

In these three years, we can think that SDN is mainly divided into two camps, the "software" represented by Vmware and the "hardware" represented by Cisco. " The SDN of "software" means that the SDN solution does not require hardware, SDN can be implemented using traditional or existing network hardware, and all the required functions are implemented in software, which is the current development trend. " The SDN of "hardware" means that the SDN architecture needs to use special SDN hardware to support and realize the function.

For Cisco's ACI scheme, there are several parts: one is the controller APIC, the second is some monitoring and other applications on the controller, and the third is Nexus 9K. Nexus 9K is the real place to achieve SDN network functions, these switches are controlled by the controller to achieve SDN network forwarding and other functions, up we can through different choreography to virtual different network environment.

This scheme still sells Cisco hardware boxes. On the Vmware side, vmware's nsx architecture is similar to that of ACI, with a three-tier architecture, and there is no difference in how to use it. The essential difference is that vmware can be implemented to support any network environment, while Cisco's ACI must be based on its own controllers APIC and Nexus9K. Vmware relies on this advantage to achieve NSX $1 billion in revenue in just a few years.

At this time, IBM also attaches great importance to SDN. The following figure shows the SDN planning and layout of IBM. The G-wan of this Google is very similar, and the problems to be solved are also very similar, but the services provided are different. IBM acquired SoftLayer, and the network connections between data centers are more or less the same as those of google. When Google was doing G-wan in 2011, he was looking for Taiwan OEM to design the hardware. Later, when IBM was doing SoftLayer, the controllers and switches directly purchased Cisco's SDN-supporting equipment, so SDN to the end, with the passage of time, network equipment manufacturers will participate in the construction of SDN network. After the completion of the later IBM SoftLayer, it now produces billions of dollars a year, while the operation and maintenance team is no more than 40 people, and it is still the same today.

However, the first commercial SDN system in China has achieved a lot of work, as shown in the following figure:

In addition to being able to access computing, storage and other resources, users can set up their own network in the data center. No matter whether you are in a physical or virtual environment, users can configure the firewall themselves. The layer 2, layer 3 and tunnel connections of colleagues across the data center are all defined by the users themselves. At the same time, it also receives a lot of vendor products and delivers them to users as part of the data center.

At this stage, each big factory is playing by itself, and there are not many commercial cases of SDN that really hit the ground. The real SDN blowout occurred in the three years from 2015 to 2017, and a large number of deployments took place in these years. At this stage, the following events mainly occurred:

To sum up, the whole industrial development process of SDN is basically sorted out clearly. The period from 2009 to 2011 is the initial stage of SDN, and the period from 2012 to 2014 is the layout stage. From 2015 to the next few years, SDN will gradually generate value.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.