Shulou(Shulou.com)06/03 Report--

I'm sorry, this article is a little long and needs to be read for 10-15 minutes. The last article got a lot of feedback from friends, soft and hard SDN has its own advantages and characteristics, for the previous text, there are two feedback. The first is a written error. In my article, it is mentioned that "as far as I know, the solutions of domestic AWS,Azure and GCP, domestic BAT and software SDN are still in the majority." there is a clerical error here that domestic should be changed to foreign, and some friends have mentioned that these Internet enterprises should all be software SDN solutions. This understanding is adopted here, aiming at the cloud parts, such as business opening and migration. The establishment of security features should be realized by software SDN. For cloud infrastructure, the initial deployment, automatic configuration, and monitoring of multi-vendor hardware switches should be completed by software automation. I understand this as one of the functions provided by hardware SDN, which is why I say that software SDN solutions are in the majority, but not all of them. The second is a message from a netizen named "Erma". The vigorous development of cloud computing and iterative technology renewal not only require the team to continuously update the knowledge system, but also need to adjust the organizational structure and management model accordingly, but also need to change their working thinking. In the past, various functional teams need to break through the situation of self-cleaning, such as the devops team and google's SRE, to break down the barriers of the original shaft functional teams and improve the efficiency of team cooperation.

I think this kind of thinking is very meaningful, the change of the cloud, technology is often just a tool, more may be needed outside the technology.

All right, back to the point, again, this article starts with the article by Daniel manufacturers, and emphasizes once again that this article is definitely not aimed at or negates specific manufacturers, or the previous sentence, which extends from the original text and gives us another perspective to look at it. Mountains have southern slopes and northern slopes, while meandering mountains look at steep peaks on the side.

The original link is as follows, I suggest you take a patient look at it. DCN academic school, data center network automatic deployment, how to choose soft and hard SDN?

The reference of the first half of this article is the first three levels of the original text, with the following links to review the old and learn the new. Technical talk-re-talk on soft and hard SDN (1)

In the second half, we start with performance.

● talks about performance

The original text proposes:

Because the hard SDN is processed on the switch hardware and even on the chip, the performance is indeed much better than the soft SDN. Soft SDN-dependent vSwitch is also making continuous efforts in this direction, such as "surround Wei to save Zhao", such as overlaying DPDK, unloading intelligent network card, etc., the performance problem has indeed been alleviated, and the data of various manufacturers have been generally raised to more than 10 gigabytes, but there is still a big gap compared with hard SDN capability. There is no industry consensus on whether some unloading technologies can be used for commercial use on a large scale, some are still in the innovative experimental stage, and the introduction of these auxiliary technologies expands the supporting boundary and complexity of soft SDN, of course, it also increases the cost.

We must be sure that there must be a gap between the software SDN and the hardware SDN purely in terms of performance, but whether the hardware is "far better" than the SDN may need to be considered by data. Take openstack as an example, why did you use hardware SDN in the early days? it is true that the performance of OVS itself is not satisfactory, and the hardware can effectively solve this bottleneck. However, times have changed, various technologies are similar to DPDK,SR-IOV, and the rise of intelligent network cards, software SDN has never stopped the pursuit of performance, while retaining the original flexibility, strive to meet the needs of the actual business, and objectively speaking, compared with a year or two ago, commercial cases are no longer limited to sparks, if according to previous standards, hardware SDN performance is 5 stars high praise, software performance is one star Now it is possible to reach 3.5 or even 4 stars, which has been illustrated by many network cards, system vendors' test indicators, and commercial cases of public clouds.

More importantly, the choice of any technology is a comprehensive consideration and the result of a balance of performance and flexibility. in the process of going to the cloud, we saw that some cases migrated from software to hardware, and some cases migrated from hardware to software. there are also cases where hardware and software are constantly implemented alternately, which precisely shows that both hardware and software are constantly pursuing progress and perfection at the level of performance. Like a marathon course, although there is a short gap, but all in the process of catching up with each other.

● openness and integration

It is mentioned in the article that there are many market segments in the enterprise market, and different enterprises have different preferences or experiences for different components, including server manufacturers and models, virtualization platforms, cloud platforms, and so on. Which of the two can better match this demand for open integration? Because the soft SDN depends on the server vSwitch, it needs to be equipped with underlying platform software, or even match the virtual server model support. Due to the differences of different virtual layer software, a single vSwitch product can not match all virtual layer platforms and needs independent supporting products. Hard SDN avoids these locks. Because the control point is in the hardware switch, it can generally adapt to the IT resource ecology connected to the lower layer, such as servers of different manufacturers and different virtualization platforms, as well as cloud platform products.

Back to my point of view, if we follow the logic of the overall planning, the cloud platform is the overall scheduling-computing, storage and network resources, and the virtualization platform hosts these implementations. SDN is between the cloud platform and the virtualization platform. The upper and lower needs can communicate effectively with the cloud platform, and the lower needs and virtualization can be accurately realized. From the point of view of the cloud platform, no matter hardware or software SDN integration is completed by API docking. There is no difference in implementation between the two, but for virtualized platforms such as openstack, or the integration of K8s or vmware, the location of hardware and software implementation is different.

Soft SDN is shown in the following figure: the goal of soft SDN is to virtualize the unified deployment of the network layer to achieve consistent network connectivity and management, which means that if you want to implement software SDN, you need to have some understanding of the adaptation of software SDN (mainly operating system and kernel, virtualization platform), and also consider whether to support mainstream platforms.

The hardware SDN is shown in the following figure: the goal of hard SDN is to keep the original virtualized network layer unchanged and to achieve network connectivity and management through hardware. And this means that if it is implemented in hardware, the unity is abandoned in the part of the virtualized network layer, and the original deployment is still retained, and it is easier to adapt to the underlying virtualization through the hardware switch. However, it does not mean that the hardware switch can adapt to all platforms, and the condition for adapting to all platforms is that the virtual network layer does not do any SDN action and can directly transmit traffic, while the virtual network layer is actually a relatively reasonable location to implement network policies. As mentioned earlier, the goal of software SDN is to achieve unity in the virtualized network layer, and its technical route must first meet most of the cloud platforms, and then consider other virtualization methods, such as VMware. We need to consider that it is a privatized platform, how the software SDN adapts to it, and what way it is compatible with it, which is the same as the hardware SDN. Therefore, when it comes to openness and integration, the goals of both are the same, and each has its own way to achieve it. From my current understanding, each has its own advantages, soft or hard, each has its own way. But back to the "lock" thing, let the software SDN carry the "lock" this pot, may need to rethink. It was mentioned earlier that the control point of the hardware SDN is on the hardware switch, so we are asking questions. Whose hardware switch is it? Referring to the implementation diagram of the previous hardware SDN, we assume that there are three virtualization platforms, V1PowerV2 and V3, respectively. Suppose the hardware manufacturer is an A, and all SDN designs correspond to V1-V3 and A. there is no problem with this logic, but if the business scale increases, we have two options for the brand of the hardware switch:

Still choose home A, which can ensure the consistency of SDN pairs.

Choose to add family B, then bring complex permutation and combination problems, V1-V3 not only need to correspond to the SDN of family A, but also need to correspond to the SDN of family B. if you add a family C, there will be more combinations.

The resulting topics include:

Can A's SDN software be compatible with B's hardware?

Can the overall SDN of home A be docked with the overall SDN of family B on the network side?

In the implementation of SDN business logic, can other network functional devices (such as firewalls, routers) be multi-brands?

These may not be a problem, because I said that each family's plan has its own advantages, using a variety of technologies to present the customer's business logic through hardware SDN, just like house decoration, if you want to be simple and easy, then find a decoration company, all things will be handed over to them to take care of, or directly buy a well-decorated house, there will be both advantages and regrets.

These may all be problems. The inherent gene of hardware SDN is that software and hardware are closely integrated, which determines that you cannot make flexible choices on hardware. We do not evaluate commercial interests and manufacturers' strategies. At least in terms of the logic of technical implementation, hardware SDN has such a status quo, which we should not and need not avoid. However, my point of view is still very "moderate". Is there a more suitable solution for "software" and "hardware"? I know the ideal is full, but we should not give up the pursuit of beautiful things because of the realistic bone feeling. The last article mentioned the foreign cloud enterprises represented by AWS and the domestic cloud enterprises represented by Aliyun. In their public cloud design, soft SDN is responsible for cloud network, and hard SDN is responsible for device management (and it is a multi-vendor model), which can be used as a reference for our architecture design. I booked a book entitled "the Cornerstone of Enterprise Digitalization" on JD.com. I hope to learn more from Ali's description. As a reference for later articles.

● costs, as well as scenarios

Many people in the industry believe that the two key points in the dispute between soft and hard SDN are reliability and cost. In the final analysis, performance problems can also be seen as part of the cost. In this race, on one side, soft SDN replaces network resources through server costs, including the cost of performance overlay and scale management. On the other hand, the hardware capabilities of network equipment vendors improve and quickly digest the cost of hard SDN. Which efficiency curve runs faster will have more advantages. At this point, hard SDN has obvious advantages at present. From the information obtained from the actual project, the hard SDN switch has basically absorbed the cost of SDN, and the price of non-SDN switch is basically the same, thanks to the rapid development of chip integration capability in recent years. In contrast, the large number of soft SDN,vSwitch plus the annual cost of operation and maintenance, the price is not cheap.

The description of hardware SDN here is very accurate, especially mentioned that with the improvement of chip capability, I will extend this topic and sum up in a few words.

The cost of 10GE and 25GE optical modules is reduced, resulting in a reduction in the overall network cost on the server side and switch side.

From a small technical point analysis, VxLAN distributed gateway has become a standard switch, the rapid iteration of the chip, so that the lowest access layer switch can have almost all the functions of SDN.

The so-called SDN switch and non-SDN switch is the ability to support the function. With the passage of time, the two will tend to be consistent, just like the difference between the so-called layer 2 switch and layer 3 switch in the past, but the price difference between SDN and non-SDN switch is objective, whether it is basically the same and how much worse it is, it depends on the pricing of the market.

In terms of analytical cost, the cost includes not only the cost of product purchase, but also the cost of learning, the cost of maintenance, and the cost of product replacement. One by one, between hardware and software, who may be expensive and who is cheap, even if the time is extended, I think it really needs a specific analysis of specific problems. What is the most direct way to consider the cost? Return may be a consideration, that is, ROI (return on investment). In an enterprise's IT investment, ROI is relatively difficult to evaluate, because IT has always been a spending department, how to evaluate the direct economic benefits? This topic is too big again. In my opinion, in addition to saying that the overall income of the enterprise is included in the investment return of IT, there may also be the following indicators, such as whether the cost of IT personnel increases linearly with the increase in the scale of IT, and if it does not increase, it is also an invisible income, which means saving money without spending too much money. For example, compared with the IT process of previous business (here we are talking about the implementation of traditional business through IT business, such as new products, new systems, etc.), is the efficiency improved or decreased? Is it more flexible or more complicated? In terms of simplicity, IT personnel have done this, do business staff feel "comfortable" in the end, and do their own operations and maintenance feel "relaxed"? These invisible things can be used as some criteria for cost considerations in practice. From the topic of cost back to the consideration of the scene, the "anchoring" of this scene mentioned in the article needs our further reflection. The original article mentioned: but specific to different scenarios, soft SDN shows its flexible opportunities to display its talents. For example, in a traditional business data center that is no longer growing, customers have no intention of investing too much in network transformation, while the old switch does not support SDN, so it will be a good choice for software SDN to replace SDN capabilities with server resources. After all, there is no need to make a big fight with a business that is not growing. On the other hand, if the business is expected to continue to grow, it is still recommended to use hard SDN to build or gradually transform, the longer the time, the greater the return.

In this statement, it is recommended to use software SDN to replace SDN capabilities in the traditional data center scenario, but as mentioned earlier, hard SDN has a significant cost advantage, so why is soft SDN recommended in the old data center scenario? After all, the matter of SDN, regardless of software or hardware, may be a process of "retrofitting" the old data center. The previous article did not mention the cost reference in the traditional data center and the emerging data center, but why should the soft SDN be framed within the "fence" of the traditional data center later?

If it were me, I might have to put it another way: for traditional business data centers that are no longer growing, the business is fixed, customers have no intention of investing too much in network transformation, and old switches do not support SDN (in fact, there seem to be fewer and fewer such data centers these years). Instead of making great efforts to transform, it is better to keep the same and maintain the connectivity and logic of the original business. When the construction of the new data center is completed, migrate the business and slowly complete the replacement of the new and the old.

In the construction of the new data center, resources are invested in research and learning. After understanding their own capability radius, determining the ideas of business development, the choice of cloud platform, and the direction of technology development, comprehensive consideration is given to the various elements mentioned in the article, such as performance and flexibility, expansibility and stability, openness and integration, cost and benefit. I believe that there are great differences in industries and technologies, but in the ever-changing world, there is always a universal truth to follow, there is always conventional logic to analyze, there are always cases of success or failure for reference, and we have come to the conclusion that hardware SDN or software SDN is not the mainstream. But whether this answer can help you meet the requirements, solve problems, and achieve growth in your actual business.

So if you ask me what the answer is, I'm sorry, it's really difficult for me to tell you directly, because my answer may not be yours, and I keep learning and overturning my own conclusion. During this journey, I saw the demand of Internet companies for performance, the pursuit of flexibility, the persistence of the financial industry for stability and the desire for innovation, and the thinking of the manufacturing industry on the Internet of things and the industrial Internet. I also saw the determination of the government and the public to reshape their own business IT. Digital medical care, new retail, self-driving, 5G, a variety of new business scenarios have brought all kinds of possibilities for the digital transformation of enterprises, among which, there are losses and gains, persistence and innovation. If you let me take the road of SDN, we might as well find an opportunity to sit down together. Have a chat, have a look. PS: alas, there are so many words here that many people no longer like to read Wechat articles with more than 3k words, but I don't bother to divide them for the sake of division. The text is also a record, and it doesn't matter how many words it is from the record itself. The next issue plans to write some technical articles, but also ask you to supervise more to avoid a bad end.

TF Live? KK/ Jianxun: cloudy, SDN, and the theory of evolution of Internet workers

Technical talk-re-talk on soft and hard SDN (1)

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.