Shulou(Shulou.com)06/02 Report--

About the author: Zhang Xin, one of the first virtualization developers in the world, is the main author of system Virtualization. In 2010, he went to Silicon Valley to join IaaS startup Cloud.com. He is the core architect of CloudStack, responsible for Oracle VM,Barematel, Baremetal VPC and other core functions. After that, he joined the software giant Citrix with the merger and acquisition of Cloud.com and continued to engage in the research and development of the core technology of CloudStack. He returned to China in 2015 to establish the open source IaaS project-ZStack.

Text

For a long time, private cloud manufacturers have been shrouded in a dark cloud: is the private cloud a false proposition, and will the world eventually be ruled by the public cloud? However plausible, when Amazon unveiled its public cloud service AWS to the world on March 14, 2006, the wheel of IT history ran into the public cloud era, and traditional IT manufacturers had to face the barbarian outside AWS. AWS CTO Werner Vogels made his claim on private cloud in 2010. Vogels said the so-called private cloud is just a sales tactic, a term coined by IT vendors to keep their customers in the dark. Vogels believes that the private cloud is "false cloud" and its purpose is to get customers to buy more hardware to "build your own cloud". Although there is a logical contradiction between "false cloud" and "build your own cloud", this does not prevent AWS from showing that private cloud is just a lifesaver created by IT vendors from high and cannot be called "cloud" at all. I have been asked this question many times over the past few years, and the core of the question is whether the public cloud will eventually rule the world.

What's it like in a world with only public clouds?

What would it be like if the world's IT infrastructure was monopolized by several public clouds? First of all, Intel will become a vassal, with the exception of producing CPU for the shrinking PC market, all its server chips can only be sold to the public cloud, thus completely losing its bargaining power. Public cloud will also use CPU of other architectures (such as ARM) to suppress the price of Intel. AWS released an EC2 instance based on ARM chip Gravtion at the re:Invent 2018 Cloud Conference.

Hardware companies such as DELL, HP and Cisco will fail or be acquired by the public cloud. All computing, storage, and networking capabilities are concentrated in the hands of public clouds, and no one will buy their hardware. The public cloud will not buy commercial hardware, but will build it itself. Do not have to wait until the future, now that public cloud / Internet giants have made extensive use of white-brand hardware, HP is no longer in the server business of public cloud / Internet companies, which is really unprofitable.

Basic software companies such as Microsoft and Oracle will die (Microsoft can rule the world again through Azure). The public cloud will replace all existing basic software with its own products, providing its own operating system, database, and all PaaS. They have already begun to do so, and will become more and more successful.

Application software vendors have to live on the face of the public cloud. If it is profitable enough, the public cloud will not hesitate to enter your field, relying on the huge technological ecology to create similar products and depress market prices. The giant has no boundaries, and AWS launched the WorkSpaces cloud desktop product in 2014.

In such a future, IT innovation is likely to stagnate. There is still competition between public clouds, but the IT market is so big that a few markets that share trillions of dollars can easily lead to a tacit monopoly among the giants.

This is like alarmist talk, after all, it is too different from the IT world that we are familiar with. But don't forget that in the mainframe era a few decades ago, a portable mobile computer (mobile phone) for everyone was more like a fantasy. And now it's all come true.

It's not sexy without SLA,AWS Outposts.

"We can rule the world," said AWS in 2010. Just eight years later, at the re:Invent Cloud Conference in 2018, AWS unveiled Outposts, a hybrid cloud product that is not sexy in terms of product form and business model. Outposts is based on AWS's own hardware (a server based on Nitro architecture, similar to Aliyun's Dragon Server), which can be deployed in customers' own data centers. AWS is responsible for installation and deployment and provides later upgrade operation and maintenance services. Customers can use popular products on AWS public cloud (such as EC2, ECS, and will include PaaS products such as RDS) in their own data centers. Outposts offers both VMWare-compatible and AWS-compatible modes, and customers can either use the VMWare control plane they are familiar with or write programs based on AWS API to seamlessly deploy applications in local data centers and AWS public clouds. In addition to the initial use of its own hardware, AWS said that in the future, it will not rule out cooperation with HP, DELL and other manufacturers to let Outposts run on the hardware of these manufacturers.

Outposts didn't invent anything new. In terms of product form, Azure Stack does the same thing. Pay-per-use rental model, some domestic manufacturers do more thoroughly, providing full leasing from rack to virtual machine application package. Compared with the public cloud, Outposts lacks a core: ultra-high SLA (Service-Level Agreement, service level agreement, suppliers' commitment to the quality of customer service, there will be corresponding compensation for failure to meet the quality of service). In essence, the public cloud sells "the right to use resources + SLA". Traditional IT vendors actually offer SLA, but the content terms are very different from those of the public cloud. SLA became popular in the late 1990s or early 2000. Early IT manufacturers provided after-sales service in the form of product warranty. As customers pay more and more attention to the online time of their business, IT manufacturers find that service quality assurance can become an additional source of income, so they sign a promise with customers through SLA to repair or replace goods within a limited period of time after a product failure. Public cloud improves the online time-related SLA of business to a level that traditional IT vendors cannot achieve, due to the control of IT infrastructure, from computer room, electricity, network, hardware to software, so it can provide minute-level SLA. For example, the monthly online time ratio (Monthly Uptime Percentage) of AWS's EC2 (virtual machine) and EBS (block storage) SLA is 99.99%, which is equivalent to a maximum of 4 minutes and 23 seconds of unavailable time per month. 10% and 30% consumption compensation will be made if the SLA is not reached. IT manufacturers cannot do such SLA because their products are only part of the overall IT architecture, and there are too many third-party factors that may affect SLA, hardware failure, power outage in the computer room, misoperation of operation and maintenance, and so on. The ultra-high SLA is very attractive to customers who value business online time. This is especially true for Internet companies. Imagine what would happen if applications like Wechat and Alipay were shut down for several days because of an IT infrastructure failure. Although Wechat and Alipay can build infrastructure similar to AWS, most enterprises with high requirements for SLA are unable to build such infrastructure on their own, and public cloud is a good choice for them.

Outposts cannot provide SLA for public clouds. Its product form makes it like ordinary IT products, but only part of the customer's overall IT architecture, and there are too many factors beyond the control of AWS, such as a massive power outage caused by a blizzard in the customer's data center. Even if AWS draws up SLA for Outposts in the future, its terms will only be similar to those of IT vendors, such as after-sales service response time, service unavailability caused by product upgrade or hardware replacement, and so on. Although the media uses many new concepts to describe Outposts, just as Vogels accused traditional IT manufacturers of adding the word "cloud" to any product eight years ago, it is just a kind of "sales pitch". It is true that Outposts has brought great value to customers by localizing some of AWS's public cloud products, but this value comes from the success of AWS public cloud. Azure and Aliyun are doing the same thing. In short, the public cloud brings great innovation to IT, but once its product loses SLA, it is no longer sexy.

Outposts is AWS's compromise to the material world.

Most Chinese and foreign media regard AWS's motivation for launching Outposts as a compromise, and AWS itself admits that customers want to have the same functionality and experience as AWS public cloud in their data centers. The word compromise means not that innovators are weak but that these customers are not. AWS has the same status as Apple's innovator in the field of mobile phones in TO B's world, and people always have unlimited expectations for subversive innovators to transform the old world without compromise. But AWS compromised, not to CIO, which has a large IT budget, but to the material world in which we live.

Although IT has helped improve the efficiency of all industries, there are still a large number of production activities in the world that have not been completely replaced by IT. Agricultural machinery is still growing grain, mining machines are still digging mines, and drilling for oil is still drilling. in short, in addition to trillions of dollars in IT spending, hundreds of trillions of dollars have been invested in other production activities. Going back to cloud computing, will there be only public clouds in the future IT infrastructure to assist production activities? AWS gives a negative answer. There are reasons for complexity, but it can be roughly attributed to the complexity of heterogeneous IT, the localization of computing / storage capacity, and the unreliability of the network.

Complexity of heterogeneous IT

Most production systems are not designed around IT, they predate IT and are now modified by IT, such as installing a large number of sensors to collect data for steelmaking equipment and robotic arms for the production line. The complexity of these modifications creates heterogeneous IT. If there is only public cloud in the world, it means that all IT have to be connected to public cloud, which is not in line with the business model of public cloud. The public cloud has grown rapidly by providing customers with standard IT models, such as AWS's earliest classic network and later VPC, to achieve scale. When the customer's IT is too heterogeneous to be directly integrated into the public cloud IT model, there are only two choices: one is to expect the customer's IT system to be modified to adapt to the public cloud; the other is to provide a private cloud to adapt to the customer's IT system. The first path doesn't work right now, regardless of how powerful traditional forces are, and in terms of volume alone, it's hard for a 100-billion-dollar public cloud to require trillions of dollars of industry to change for it. The only alternative is to provide a private cloud to accommodate the customer's IT system. Opponents may say that AWS Outposts's output is still a public cloud product, with no special changes, which will be left for later analysis of the Outposts product form.

Localization of computing / storage capacity and network unreliability

Even if heterogeneous IT can be connected to the public cloud, it still faces the limitations of current IT technology. We made up an AI system, which can control steelmaking equipment in real time according to the sensor data of steelmaking equipment. So where is the sensor data stored? where does the AI system run (where is the computing power that needs to be relied on)? Assuming that the steel plant is remote, the steelmaking equipment must be connected through the network to a public cloud data center thousands of kilometers away, transmit the data to the AI system and wait for the results of the analysis before doing the next step. Such an IT system is difficult to be confident because it runs on an unreliable network. Compared with the total amount of data in human society (rumored to be 100ZB, about 10 to the 14th power GB), the current network is definitely a low-bandwidth, high-latency and unreliable system. In order to ensure the bandwidth and delay of the above steel mill network, it is necessary to connect data centers thousands of kilometers away through dedicated lines, and there must be redundancy, otherwise it is impossible to prevent production disruptions caused by broken lines of the construction team's forklifts. This is undoubtedly a huge investment. If the sensor of the steelmaking system is replaced with a camera and the AI system is required to make low-delay image recognition, the cost of network construction is even more unimaginable. The solution is to localize the computing / storage capacity, for example, to build a computer room within 500m of the steel plant to let the AI system run nearby, and the network problem will be solved. For cloud computing manufacturers, exporting a private / hybrid cloud system to the computer room of a steel mill is simpler and more reliable than designing a system that connects to the public cloud across thousands of kilometers, which is the limitation of current IT technology.

The example of manufacturing may sound distant, but even in the day-to-day office, public clouds can have the problem of IT access. When AWS launched WorkSpaces cloud desktop products in March 2014, my first reaction was to think about how AWS cloud desktops connect to office printers, scanners and videoconferencing systems, because the main scene of cloud desktops is daily work, which inevitably has to be docked with various devices in the office. Later, I found the documentation for the printer on the AWS website, but I also saw a lot of complaints on the AWS forum that the printer didn't work. There are many such examples, such as bank U shield, intelligent projector, etc., in short, it is not so easy to access local devices in the public cloud.

If we live in a virtual world without food, clothing, housing and transportation and only rely on the exchange of information (wearing a helmet to connect to the virtual world as in science fiction movies), then the public cloud can fully meet all the IT needs of production activities. I say AWS is a compromise to the physical world because the current public cloud only meets part of the IT needs of the real world. If you take a look at the innovations of Apple and AWS, you will find that Apple has wiped out almost all the established mobile phone manufacturers, Nokia, Motorola and BlackBerry in just a few years after the launch of the IPhone. Although AWS has dealt a huge blow to traditional IT manufacturers, IBM, HP, DELL and Oracle are still strong and alive. This is because IPhone not only has great innovation, but also meets all the needs of customers for mobile phones. If IPhone cannot make calls, Nokia will still be the best mobile phone manufacturer. Although AWS, like Apple, has driven the development of new industries (mobile Internet, SaaS, etc.) and reaped huge innovation dividends, it has not met all the IT needs of the world. Traditional IT vendors survived, making private clouds that AWS called "false cloud". Now that AWS Outposts has come, blowing away the last dark cloud in the private cloud sky, there is no need to debate whether the private cloud / hybrid cloud is a false proposition, whether it is "true cloud" or "false cloud". Facing up to the complexity of the IT needs of the physical world, maybe the public cloud will meet all IT in the future, but not now.

Doing the right thing with VMWare,AWS again

Outposts is a cooperative product of AWS and VMWare. Customers can use the control panel of VMWare that they are familiar with, or they can use AWS Native mode. The first generation of products are mainly IaaS functions, providing EC2 and EBS. It is certain that the IaaS part of Outposts is almost entirely contributed by VMWare, and AWS also claims that customers can use VMWare NSX, AppDefense, vRealize and other products. From a technical point of view, there are two ways for AWS to implement Outposts, one is code modification based on AWS public cloud, and the other is to cooperate with vendors like VMWare. AWS is right to choose the second path. Public clouds and private clouds face different complexities. Public clouds put more emphasis on scale and performance, providing resources with a unified IT architecture, and basically ignoring the heterogeneity and usefulness of traditional IT. Private clouds must face heterogeneous IT forms and be compatible with a variety of legacy devices, but they do not have to be on a par with public clouds in terms of scale and performance. Outposts's target customers are enterprises that want to use AWS products in their data centers, and most of them are traditional IT architectures. It is very difficult and risky to modify AWS public cloud codes to meet customer needs, which means that AWS public cloud products have to be considered for private clouds in the subsequent development, which will not only slow down the pace of public cloud innovation, but also difficult to operate. Imagine that the kernel engineer of AWS has to consider a PCI device he has never heard of when optimizing the hot migration of virtual machines, because private cloud customers will pass it through to the virtual machines. Using VMWare to do IaaS does not have this problem. VMWare has always served traditional enterprise customers. It has made numerous trips over the past 20 years, precipitating a large number of functions needed by traditional IT. It was mentioned earlier that AWS can adapt to the traditional IT needs of enterprise customers by deploying Outposts. Opponents may think that Outposts also outputs AWS public cloud products (such as EC2, EBS) and does not make any adaptation for heterogeneous IT. Yes, AWS didn't do it himself. VMWare helped him do it. AWS claims that Outposts can be run on a server, which cannot be done without VMWare. Stuffing all the control and data planes of the AWS public cloud on a server and running a customer's own business is tantamount to stuffing an elephant into a refrigerator.

In business strategy, partnering with VMWare is also the right choice. In addition to its own product advantages, VMWare is the most powerful sales network all over the world. Foreign media reported that Outposts will be sold in the VMWare sales network, which undoubtedly greatly saves the time for this product to reach customers. It takes a lot of effort to make a new product quickly recognized and accepted by the market, even with the highlights of AWS. This is not only a technical problem, but also a problem of sales system, supply chain and after-sales service system. Amazon itself is an Internet company and prefers to sell online, while VMWare is good at selling offline. In this cooperation, no matter how you look at AWS, you will benefit. It is hard to predict the misfortune and good fortune of VMWare. Once opening its own sales network to allow AWS to reach end customers, it is difficult to say that AWS will not set up a special department to completely replace VMWare with its own technology and establish a suitable offline sales system after the product has been established. In addition, AWS is not the only VMWare partner, HP, DELL, Cisco may be waiting to welcome Outposts into their sales channels.

Contrary to AWS's technical approach, Azure Stack chose to use its public cloud code to transform hybrid cloud products that can run on several servers. Microsoft has always had the urge to unify code base, trying to build an operating system for PC and mobile phones with the same code as Windows, which ended up withdrawing from the smartphone market. Although it is impossible to tell whether Azure Stack's technical route is successful, it will always face the problem that the same set of code has to serve two very different markets. Internal technical team coordination will also encounter challenges. A strong public cloud team will inevitably lead to a lack of timely response to many technical requirements of Azure Stack, while a strong Azure Stack team will slow down the pace of innovation of public cloud teams, or even drag them into the quagmire of traditional IT.

I personally agree with AWS's approach, ensuring product experience consistency is the core, which only needs to be compatible with the public cloud on the API of the data plane / control plane of Outposts. As for one set of code or two sets of code is not important, one is that customers do not care, and the other is that the market scale of both public and private clouds is very large, so it is worthwhile to operate two technical teams to serve different markets. AWS also mentioned that they do not intend to export all AWS products through Outposts and acknowledge that public and private clouds have different delivery models (Our initial goal is not to re-create all of AWS in Outposts-they're kind of different delivery models. But there are some really basic components that we're hearing consistently that are wanted on-premises). This reflects that AWS's positioning of Outposts is very clear, and it is impossible to deliver all AWS products to customer data centers in a small-scale product form. In fact, enterprise customers can not afford and maintain the entire AWS public cloud, most enterprises only need a few basic functions of AWS, which is the purpose of Outposts. Of course, there are customers who need the entire AWS public cloud, such as the U.S. Department of Defense's $10 billion order, so AWS will naturally bring all its products to the customer's data center, not with Outposts.

The Future of the Public Cloud: rule the World or perish

In the future, the public cloud will either dominate all IT infrastructure or disappear, and there will no longer be the word cloud computing. This does not depend on the strength of current IT practitioners, but on physicists' major breakthroughs in basic physics to subvert the existing computer system. Among the several problems of public cloud access to traditional IT, as long as the network problem is solved, others can be solved. At present, the low-bandwidth, high-latency and unreliable network has greatly hindered the pace of the public cloud to dominate the world. Although the network performance is constantly improving, this small pace is far from keeping pace with the growth of human production data. If basic physics creates a super network whose access is not limited by regional distance and achieves the bandwidth, latency and reliability of current CPU access memory, then the world only needs public clouds. We no longer need local computing power, and we can centralize all computing and storage in the data centers of several public cloud giants and access them over the super network. If basic physics makes a breakthrough in computing / storage capacity, assuming that computers the size of mobile phones in the future can have all the computing / storage capacity of today's AWS public cloud, then we no longer need the concept of cloud, computing can take place at any time and place, and the centralized infrastructure has no reason to exist.

The future of the world is still unpredictable, but at this time, AWS released Outposts products, and he finally entered the same dimension of traditional IT manufacturers. From dimensionality reduction to competition in the same dimension, the field of cloud computing will be more exciting in the next decade.

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.