Tang Yuanwu, Chief Engineer of Network Products Division of Xinhua Group: NFV Optimization and Enterprise Scenario Application

Top Network Network Channel August 3, Beijing Report: The 2017 Global SDNFV Technology Conference (2017.chinasdn.org) hosted by Tiandi Interconnect and the Next Generation Internet National Engineering Center was held on August 2-3, 2017. Beijing Guobin Hotel opened fiercely. In the afternoon of the 3rd, Tang Yuanwu, chief engineer of the Network Products Division of Xinhua Group, shared a wonderful speech entitled “NFV Optimization and Enterprise Scenario Application”.

Tang Yuanwu, Chief Engineer, Network Product Department, Xinhua San Group

The following is a live speech:

Tang Yuanwu: Hello everyone. I am Tang Yuanwu of Xinhua Third Product Department. The topic I shared with you next is NFV optimization and enterprise network applications. Starting this morning, it may be more important for the operators to share the application and trends of the entire NFV in the network reconfiguration of operators. From the perspective of NFV, it is possible that the operator is the core of the entire operator network reconfiguration. Today we want to say a different perspective, from another perspective, what value or what kind of scenario NFV has for enterprise network applications. Because everyone knows that Xinhua San may be our position for Xinhua San for the enterprise network.

Let's take a look at our SDN and NFV from a relatively microscopic perspective, because no matter who you are from the speeches of the guests this morning and the actual application of the scene, SDN and NFV are inseparable from anyone. From the network infrastructure dimension or from the original intention of SDN and NFV, we believe that SDN was originally designed to solve the following problems of the OS seven-layer network layer, because we know that the transfer control problem is more directed to the network layer below, control and forwarding. Separation. However, with the application of the entire SDN in various industries, SDN has a tendency to go up.

That is to say, we often talk about the separation of control and forwarding, but in fact we really want to achieve it, or we drive us to realize the separation of control and forwarding. The essence comes from the upper core application. If we leave the upper application, we simply talk about forwarding and control separation. So obvious, or from the current application, if you leave the application, it seems that the driving force is insufficient. This is also the theme that Xinhua San has been running through in the enterprise network SDN solution, which is called application driver.

For example, our application-driven data center, application-driven campus network, and application-driven WAN are also very important for us to consider in the application process of SDN. Similarly, for NFV, we know that the original function of network function virtualization is more than the OSI transport layer. Of course, for Huawei, the virtualization of network functions is nothing more than a platform to another platform. I think the technical difficulty itself is not very big. From the current development situation, the real restriction of NFV, in addition to the large network programming mentioned by the guests in the morning, in fact, for the network source, the performance of a single network source has become a relatively large NFV continuous development. The bottleneck.

For H3C, what is our core philosophy at NFV? Basically, it is similar to all the guests, which is the basic resource of the standard, and strives to provide the concept of our NFV service with the overall open delivery of the standard. On the one hand, we have to play the flexible definition of NFV itself from hard to soft, the characteristics of rapid deployment, and we will combine the NSDN concept, let NFV in addition to the two points mentioned just now, we have to expand the interface. This is actually very meaningful. Because the simple meaning of softness is not great. As I will mention later, as long as the resource pool is formed, I think that high reliability is a great improvement compared with the original single network source.

H3C's goal of NFV is divided into three according to the normal NFV architecture. The same goal is to follow these three parts separately: first in the NFV base layer, the first one is hardware resourceization, I think this is a commonality. . Based on this hardware resource, because of our current practice and project promotion, high-performance forwarding at the NFI layer may be an aspect of our H3C's focus on exploration or innovation.

Secondly, for the network source, that is, we have already completed the stage from soft to hard. What will happen next? We will do more practical innovations in the network source layer to meet the needs of users. It may be two dimensions. One is the original location of the network source, as well as the Docker behind, and the direction of maintenance. Also, we hope to use SDN technology to achieve on-demand creation. I think this is a more important value for everyone after resource pooling. In the above words, there are many words for the layout layer. We should say our standard, the first one is to follow the standardization. The second one can adapt to the layout of the big network. This is our goal in NFV directionality.

Specifically speaking, we talk about the actions of each level. The first one, NFV is a bottleneck with the performance of going forward. The optimization of performance has become a matter of importance in the industry, including in the operators, that is, if we really reach us. If the desired NFV can reduce our cost, in fact, the rate of forwarding in a single network source or unit space is actually a very critical factor. From the current progress, from the NFV software virtualization, to the NFV standard forwarding, basically all of the use of NFV in the communications field is achieved using the SROV network card.

At present, SROV can basically approach the efficiency of hardware forwarding. In the next step, the current rate of access to the physical machine can be seen from a single network card and virtual machine. If we want to expand more network cards, in fact, the NFV computing platform we originally envisioned, this ability is actually found to be insufficient, that is, our CPU cannot support the expansion of large-scale network cards. On the contrary, to solve the problem of the network card, the CPU becomes a bottleneck. In the next step, we may think that we will create the so-called intelligent network card era, that is, by uninstalling the CPU offloading, so that the entire network card has a large amount of CPU processing power, so as to achieve NFV performance improvement.

The second dimension from the entire network source is not just our soft replacement. From the source dimension of the virtual machine to the container, to the subsequent atomic microservices, we are also from the perspective of a single network source optimization, we hope to say that the function is further split, because the resource responds quickly in this split. Finally, the carrying capacity of the entire large network resource is improved.

The following is the improvement of the NFV architecture. In fact, we will consider it as three stages. First, the soft camp, whether in the operator or the enterprise network, this action has basically been reached, but for the actual meaning of our entire network. Not necessarily has a lot of meaning. Everyone develops to the second stage, which is resource pooling, because the single soft replacement is nothing more than a change in form. Only when resources are pooled can one scale effect. In the morning, Bi also mentioned that it is the operator who has this idea, that is, when the NFV replacement reaches a percentage, the real cost effect will be reflected.

In fact, this embodiment actually requires pooling of resources from the dimension of technology or from the perspective of the whole architecture. This is the effect of scale. Secondly, it has the flexibility after pooling resources, and can meet different business hours after flexibility. Or the needs of different stages. The second phase, it should be said that it has been fully realized in the practice of operators. The third phase will go a step further. In addition to forming a large resource pool, it may be NFV cloudization, which is true network cloudization. There are several key technologies here, such as our control and forwarding for a single network source. Separation of the architecture, such as each NFV network source, we will abstract the forwarding and control platform internally. It is originally separated. Now, further, the NFV software needs to be further extracted and segmented from the control and forwarding platform.

Second, after splitting the database, the control plane may adopt the database technology. The advantage is that we may maintain a user's network in the future. Maybe we no longer maintain so many connections, but maintain one. The database, in this case, will bring even greater changes to our future network simplification. This is in terms of the complete dimensions of the entire NFV architecture.

Going back to a topic that I want to share with today, NFV is at the operator, I think this value is self-evident. What kind of value will there be in the enterprise network? Because Xinhua San has been serving the corporate network for a long time, it has always been our place in the corporate network.

From the current point of view, I think there will be three levels of change: First, rapid business deployment. That is, through the NFV soft-hard form, it will become very fast at the level of our business deployment. In fact, this is actually done in the data center, in our so-called SD-WAN wide area network, or in the campus network. In fact, we all have this feeling, compared to traditional hardware, regardless of procurement. The cycle, or deployment from our entire network configuration, is a big improvement. Second, the elastic expansion of the network source.

That is to say, it has just talked about the resource pooling and the flexibility brought by the pooling. This flexibility is not only meaningful to operators. In fact, for many enterprise networks, especially in the DCs after many enterprises have clouded, in fact, It also has a very big meaning, and there will be a PP T to share with everyone. Third, I think there is still a continuous investment protection in the enterprise network. Because, however, it is known that the equipment of the enterprise network itself is updated relatively quickly. Due to performance reasons, functional reasons and the reasons for using the environment, there may be a replacement cycle within 3-5 years.

From the perspective of NFV, after decoupling hardware and software, it is protected from hardware and software maintenance. Take a look at NFV to achieve cloud software combination access, this picture can be seen, that is to say, the cloud has become a normal state, but in the cloud has a basic feature, that is, flexibility. There used to be computational virtualization, so the computing layer can be flexible. The network is now resilient, but for public clouds or self-built private clouds, there is a very important concept that is tenants.

How to provide this kind of flexible access for each tenant in the network, especially for public cloud services, such as many users on Amazon need to access, this time secure traditional network equipment, regardless of the configuration dimension or security The dimensions of isolation are relatively difficult. Now use the NFV form, because the cloud language generates a large number of virtual machines, which is to allocate network resources to users like the allocation of computing resources, which is the value of NFV to users and public cloud operators in the cloud environment.

I have summed up three things: First, tenants are networked. It is the tenant network cloud. Second, the rapid business opening. Using the NFV form has become an action to apply for a virtual machine, but the mirror is different, and the service is opened. Of course, the release of resources is simpler, and the virtual machine can be directly killed. Third, high-intensity business isolation. As mentioned in this figure, it is possible to use a single NFV network source to provide multiple tenant services, or a tenant to provide a network source service. Of course, this is actually technically not a problem based on our own performance on the connection and our judgment on the size of the user.

So if you look at this scenario, for example, we can now deploy with NFV on Amazon. This is the first scenario, a scenario that is commonly used in cloud scenarios. The second scenario implements a flexible multi-service processing resource pool. Here I want to emphasize that at the beginning of NFV, everyone's first reaction may be stronger than calculation and weaker than forwarding. At this time, I want to deploy as long as the computing power is strong. The fact is that it is like this.

There are many scenes in our environment, such as our firewall, such as IPS, which is a common occurrence. Similarly, for our corporate network, for example, we have just mentioned in a DC, there is a virtualization scenario after the calculation, but there is still a point that is not flexible, it is safe. Like firewalls, IPS, if this thing is hardware, it is not flexible. How to make the security business bounce, in fact, if you use NFV, you can achieve the computing network security business processing we want, all pop up, so It becomes more effective for our entire operation and maintenance.

The security flexibility, we think that from the current point of view, the form of NFV, providing this kind of service is a relatively good way, with the resource pool I just mentioned, using SDN technology to achieve resource pool flexibility Expansion, in this case, we may put the network in the DC and bounce all the security. Here I summarize three points: First, based on the flexible expansion of the business. When the business volume is large, a virtual machine. Second, the same business is rapidly deployed. The third, has a similar effect, continuous protection.

In the third scenario, this scene is relatively smaller. Using NFV to build a university service platform, there is an experimental environment in the school. This environment has a characteristic that the environment of each class is different, using NFV, As with the infrastructure, as long as the virtual machine is available, the network function can be deployed at any time according to the needs, and a router is required to be installed on the router. So from this dimension, this scene is very suitable for colleges and universities to build a network experiment platform, which can be dynamically created on demand, and restore the environment with one click. There is a big improvement in both the investment dimension and the business agility.

In the last scenario, NFV is used in the In One access of the enterprise branch. This is a relatively large number of functions superimposed on the traditional communication equipment. The traditional embedded method of this overlay research is continuous application. If the NEF form is used, the hardware is standardized. This kind of upgrade becomes very simple, and the superposition of functions becomes very simple and very easy. Actually, this will be involved in our later edge calculations. I think this is also a direction.

This is probably part of the scene I shared with you today. I also shared our NFV cooperation plan. First of all, we provide our own network source, but we not only provide our own network source, but also introduce third-party network source on our architecture. To build, the network source that is introduced can also enjoy all the capabilities just mentioned. The second is to build a lab for partners, and our entire architecture can also serve this experimental platform. Finally, we will have a relatively detailed cooperation plan. We hope to provide a platform for technology and achieve a win-win situation in business.