What is the future of cloud computing? Debunking the Amazon Cloud Tech Growth Myth

This article is reproduced from the public account: CloudTech2030
The article only represents the personal opinion of the original author

I watched the live broadcast of the Amazon Cloud Technology Innovation Conference before, especially Gu Fan's keynote speech "Reconstructing the Cloud Base and Accelerating the Future". Compared with the grand venue and rich topics of re:Invent, this time I can get a better overview of the overall situation of Amazon cloud technology. Products and solutions, I happened to be thinking about the development of cloud computing in China and the United States, and this article was inspired.

Cloud computing is described as running water, convenient and fast. In the early years, there was no meter for tap water, and paying per capita was a serious waste. Only a few rural villages do this. But can you imagine that China's cloud computing has been the same as rural tap water in the past decade. End users don't save money, and service providers don't make money.

This article is a bit long, so let's start with the conclusion. Amazon e-commerce has a flywheel growth philosophy, and cloud computing business also adopts a similar growth flywheel strategy. "Active price reduction" increases customer stickiness, and "architectural innovation" customers realize business value. It seems simple and easy to understand, but it contains a profound product philosophy. Many followers cannot get what they want and cannot imitate. Let us interpret the product method of Amazon Cloud Technology and solve the mystery of cloud computing growth.

Refactoring the cloud base

Compared with the re:Invent conference at the headquarters, the Chinese version of the live broadcast is easier to understand, and the information is condensed. Finally, "it will be at the top of the mountain, and the mountains and small mountains will be seen at a glance". keynote's speech on cloud base infrastructure products allowed me to re-understand cloud computing product strategy from a more global perspective. The questions lingering in my mind gradually got answers. Why do so many users choose Amazon cloud technology, why is the scale of 62 billion US dollars, and the growth rate can reach 37%?

Problems and thinking of cloud computing industry

"The total revenue of Amazon's cloud business in 2021 is $62.2 billion, a year-on-year increase of 37%. Revenue in the fourth quarter is $17.7 billion, a year-on-year increase of 40%. The full-year operating profit is $18.5 billion, with an operating profit margin of 29%." This news, lamenting such a large scale of revenue and such a high profit margin, is it because the product is priced too high because of binding customers?

Looking at the domestic cloud computing strategy of losing money and reducing prices, shouldn't this make the market bigger?

Let’s start with the conclusion. Amazon Cloud Technology adopts smart product positioning and different pricing strategies to save money smartly for customers, while earning profits and investing in technological innovation and globalization. The above is the growth flywheel of the overall strategy, and the following is the breakdown from the product perspective.

The growth flywheel of cloud products:

Compared with the growth spurs of cloud computing vendors (part of the followers), the spurs need extra thrust to keep running, such as customer subsidies, heavy investment in customer relationships, customization, etc. This may be the reason why many manufacturers continue to lose money, but their share cannot grow.

Cloud computing creates value for customers

So far, there has been no disruptive technological invention of cloud computing, and no one has made it clear what its value will be in the future, and how big is it?

Gu Fan quoted the McKinsey Quarterly report, "By 2030, cloud computing will create over $1 trillion in value for the top 500 companies, save $430 billion in cost, and create $770 billion in value". Cost savings greatly reduces the cost of self-built IT software and hardware for customers, and at the same time, uses the easier-to-use AI and big data technologies on the cloud to create value for enterprise business.

The article is a bit long, let's use McKinsey's pyramid thinking framework, and the conclusion comes first.

The user stickiness and competitiveness of cloud products have the following three aspects:

1. Cost Savings: Redefine IT Procurement Models to Help Customers Optimize Costs and Achieve Win-Win
2. Creating value: Reconstructing the cloud base around user value, building advanced computing, products and services from the chip to the data stack
3. Globalization: Help customers globalize through globalized network and resource layout, localization compliance, and data security

Later, I mainly share my thoughts from the first two aspects.

This article focuses on 1 and 2 first.

1. Redefine IT for cost savings

Or the conclusion is the first, Amazon cloud technology saves costs for customers, not by low prices, but by product definitions that are close to the user's IT model. For example, pay-as-you-go Amazon EC2, commitment savings plans, flexible sharding containers, low-cost preemption Spot, serverless Amazon Lambda, and Amazon S3, which automatically downgrades and saves money.

For example, Amazon EC2 pays per second, although the unit price is 50% higher than the regular annual price. If your application has a high load time of less than 16 hours in one day, it is more economical to pay by the amount. Smart products do not rely on low prices, and win-win.

Traditional IT gameplay

Past: IT veterans know how the field was played in the past. The market of IT software and hardware products is complicated, there are manufacturers-channels-agents-integrators, and the agents are also divided into general/first/second generation. Manufacturers also need to build teams to manage channels, manage price systems, and different industry/regional/agent-level discount systems. They also need to prevent cross-selling. Manufacturers spend a lot of money on channel incentives and management.

From the customer's side, Party A's enterprise also needs to maintain a strong procurement team, negotiating, comparing, shortlisting, and signing orders. The intermediate links are complex and opaque, and Party A's boss is not at ease. It also needs to support the back-end clean government, compliance, and auditing. .

For example, our common enterprise-level high-end server, storage, network and security product pricing official website cannot be found, the discount range is huge, the minimum can be 10% off, or 50% off. Customers with weak bargaining power need to find multiple channels for inquiries, compare various suppliers, and reduce prices by various means. Originally, the order could be delivered in one month, but after the Duolun game, it took half a year to toss, and then multiple combinations were integrated, and then it took half a year to deliver.

Other players bind Dafa

Cloud computing is described as running water, which is convenient and fast, used on demand, and billed according to the amount. However, there was no water meter at first, and it was charged on a monthly basis, and now it is still operated in rural areas. As a result, those who drink water are too expensive, those who sell water lose money, and only those who water the fields have no problem.
But China's cloud computing has done just that in the past ten years. The main sales method is annual subscription. It looks like the discount is low and the end user is not saving money. I finally figured out that I want to change the model, but I found that the flexibility is not enough and the sales rate is not good... The growth flywheel cannot be imitated by others.

See the 3-year price adjustment notice for a cloud product:

2. What products does Amazon Cloud Technology consider for customers?

1) Product strategy of continuous price reduction

Customers choose cloud computing and comprehensively evaluate TCO, performance, reliability and security, experience and service capabilities;

If the service provider grabs the order at a low price, be careful if there are any traps, and will it be shoddy? Is it possible to kill customers after entering the store? Can ultra-low discounts last in the long run?

Amazon's cloud technology products have been reduced in price dozens of times. Compared with other players, they do not adopt a low-price strategy, but a win-win product design strategy, which is worth thinking about.

Next, we focus on analyzing the main revenue products. Enterprise IT spending IaaS generally accounts for the bulk, and the largest product is computing and storage that add up to generally more than 50%. Let's take a look at the product strategies of cloud vendors.

2) Lighter and lighter product architecture

Savings from application architecture:

Virtual Machine Amazon EC2--> Container Amazon ECS/EKS--> Amazon Fargate--> Amazon Lambda

From the form of physical servers in the past, to more and more fine-grained segmentation, the final realization only takes up resources when calling. Resource overhead can be more closely aligned with the customer's IT load curve. Reduce waste and help customers save money.

Save on pricing model:

Pay-As-You-Go-->Reserved Instance-->Saving Plan--Spot, while focusing on pay-as-you-go, gives users a variety of flexible options.

3) Stick to the pay-as-you-go model

The biggest difference between cloud computing services and traditional IT is that they are charged by subscription. Common software subscriptions are generally on a monthly or annual basis. However, the Amazon cloud service is more aggressive. The service subscription is paid per second (minimum 60 seconds), and there is no long-term binding, which just solves the customer's IT peak pain point. According to the current pricing ratio, if the daily usage time is less than 16 hours, it is more cost-effective than the annual subscription.

Table - Differences between various payment methods

Long-term adherence to billing by volume, providing a variety of flexible solutions and pricing strategies, cultivating the market's habit of using resources by volume, and incubating serverless products. It can also be seen from this speech that customer flexibility is very common. 60 million new instances are created every day, which is estimated at 4 vcpu per instance, representing 3 million servers, and over 50% of the application load on the cloud is created elastically.

In the past, hardware equipment of IT service vendors could not be subscribed by volume, and could only be purchased according to peak demand. But why don't other domestic cloud vendors follow suit? Why is this model still the main push (many service providers want to bind customers with low-cost packages for N years)

Essentially, paid package service providers need to reserve resources, capacity, and performance to ensure a high success rate for redemption when flexibility is required. There is a huge cost to support, and sufficient server resource pools are reserved; it is also necessary to build a large enough availability zone computer room and pay the cost. The region of Amazon cloud computing in the United States is generally 3-4AZ, and the number of AZs of domestic service providers often reaches 2 digits. Short-term costs are saved, but the resource pool is fragmented, which affects the customer experience of elastic expansion.

4) Preemptive instance - a new paradigm of shared resources

A New Paradigm for B2B Shared Resources, Auction of Idle Resources

With Amazon EC2 Spot Instances, you can request Amazon EC2 spare compute capacity, saving up to 90% on the price of On-Demand instances.

Calculate savings compared to On-Demand over the past 30 days

Outage Frequency represents the rate at which Spot has reclaimed capacity over the past month.

In the past, there were many peaks and valleys of customers, and it was very difficult to schedule peaks and valleys, which often led to a 30-40% limit. By using the spot method, the remaining resources could be seized at a low price.

In fact, many companies that do computing power buy this spare resource at a low price and distribute it to customers who need computing; for example, parallel computing is similar to players, and secondary sales of HPC business.

5) No sword in hand--create a new track without server

A service provider that started out as a cloud server has actually started a new serverless track. Top masters pursue the level of having no sword in their hands and having a sword in their hearts.

No matter how the infrastructure uses resources in a quantitative and flexible way, people have a scale-out water level. Generally, the application working water level is 30%, and the water level for starting expansion is 50-70%. Customer service is over.

The pioneering product Amazon Lambda is released, which pays completely according to the business load, and reduces the resource occupation and waste to zero.

Amazon Lambda charges = compute time + requests + memory + storage + concurrency.

Let's look at an official case, a take-out ordering system with peak and valley characteristics; it handles 3 million requests per month. Function Compute implements an execution time of 120 ms with 1536 MB of memory each time. After deducting 1 million free credits, it is only $20 a month;

Of course, serverless is not a panacea. It only supports scripting languages, has slow startup (50ms+), and has call limitations.

6) Amazon S3 for automatic money saving

Amazon S3 is object storage, the largest storage product by Amazon Cloud Technology. From the perspective of development, when data is generated, it is not known how to use it in the future, whether it is cold or hot data in the future. And Amazon S3 storage provides automatic grading of hot and cold, and it can be automatically downgraded if it is not used for more than 90 days;

I would like to ask a question, so actively helping customers save money, does PD not have to bear the revenue KPI?

Design a product and try to collect as much money as possible from customers. This is professional PD.

Amazon S3 is not just a product, it's a family, with multiple surviving paradigms of popularity and performance.

Amazon S3 product prices decrease as usage increases. According to my expectations, the price should continue to decrease after the capacity reaches PB. The latter should be left to the sales to apply for discounts.

The fly in the ointment is that Amazon S3's charging complexity consists of six major costs: storage pricing, request and data retrieval pricing, data transfer and transfer acceleration pricing, data management and analytics pricing, replication pricing, and processing data using Amazon S3 Object Lambda. price.

How to choose a handy weapon--summary of purchase methods

Boldly imagine that at present, large customers still have additional discounts, and there is additional room for bargaining. In the future, the industrial Internet will automatically generate tiered discounts according to the purchase volume, which will be more transparent. I hope that some cloud manufacturers will take this step as soon as possible.

Taking the initiative to help customers save money, this may be the reason why they dare to charge by volume, and the customer stickiness is getting stronger and stronger.

Seven Weapons of Cost Management and Optimization Tools

Good product design needs to be financially friendly, and enterprise IT departments have cost-cutting KPIs every year. Amazon cloud technology has so many tool combinations and powerful functions. Many bills can also be exported to excel for secondary structure processing.

Rich toolkit:

a. Amazon Pricing Calculator to estimate the cost of cloud usage in the Amazon Cloud Technology China region
b. Built-in product: The purchase interface has built-in suggestions for various options, such as Saving Plan, input requirements, and automatically give suggestions.
c. Cloud Financial Management (CFM) enables organizations to align their processes to achieve maximum business value and financial success while optimizing costs on Amazon Cloud Technologies.
d. Amazon Cost Explorer View and analyze your costs and usage. The tool provides default reports that help you visualize costs and usage (e.g. accounts, services) or resource levels (e.g. Amazon EC2 instance IDs)
e. Amazon Trusted Advisor, an online tool that helps you configure resources according to best practices in five areas: cost optimization, performance, security, fault tolerance, and throttling. Regular use of Amazon Trusted Advisor is recommended to best maintain your solution.
f. Amazon Compute Optimizer recommends the best instance type and size for your workload to reduce costs and improve performance, and uses machine learning to analyze historical utilization metrics. Over-provisioning resources can lead to unnecessary infrastructure costs, while under-resources can lead to poor application performance. Helps you choose the best configuration for three types of resources based on your utilization data: Amazon EC2, Amazon EBS, Amazon Lambda.
g. Amazon QuickSight cost visualization solution

Technological innovation revolves around customer value

Bottom-up fundamental technology innovation-chips

Amazon Cloud has been deeply involved in customer IT for many years, designing chips for cloud native scenarios, starting from reducing virtualization loss, and gradually taking over customer general load and ML load:

Amazon Nitro: It is a smart network card that supports virtualization, bringing the concept of DPU to fire. Rapid innovation, performance benefits to customers
Amazon Graviton: The CPU based on the ARM architecture has now developed to a level that can surpass x86 in terms of ecology and performance.
Amazon Trainum and Amazon Inferentia: ML training and inference chips, respectively, are more cost-effective than NVIDIA GPUs.

Amazon Nitro Accelerates Application Performance

Amazon Cloud Technology launched the Amazon EC2 virtual host service as early as 2006. The only virtualization technology available in the early days was VMware commercial software and the only open source Xen technology. KVM was just born and immature.

Xen's virtualization adopts full virtualization technology, that is, relying on software to virtualize memory, IO, peripherals and other equipment, and the CPU performance loss is 20-30%. Mainstream cloud service providers have switched to KVM. In 2013, Annapurna labs was acquired. Still suffering from inefficient old technology. After being beaten by VM performance for two years, Amazon Nitro and Amazon Nitro hypervisor were finally released in 2017 to optimize virtualized workloads.

Table - Amazon EC2 Virtualization Types

In the past, KVM software virtualization implemented SRIOV, but vNIC and vBlock devices still required a large number of cores to process device IO, competing with application loads for CPU resources. The load ratio is about 10-20%. Using Amazon Nitro hardware virtualization technology to help customer applications achieve performance improvements, which means lower resource requirements and lower costs:

Memcahed: Amazon Nitro leads 9-26%
Nginx: 11-20% ahead
MySQL: 6% ahead

Amazon Nitro SSD optimizes storage latency: from 0.08ms to 0.02ms; databases and ML require lower latency to improve application response speed. In the past, commercial solutions should be used, but now Amazon Nitro accelerates, realizes SSD virtualization, unified monitoring and management, and reduces latency.

Amazon EC2 rolls out new instances quickly

Amazon cloud technology is mainly provided to customers in the form of services. The background technology is iterated, customers do not need to perceive, and a product series type can converge to 1-3 forms. However, with different computing products, customer applications will perceive differences in CPU platforms, and different applications have different requirements, which leads to the rapid explosion of Amazon EC2 product instance series. For example, CPUs are divided into Intel, AMD, and ARM, and memory is divided into 1:2, 1:4, 1:8 and 1:16, accelerators are divided into inference, training, FPGA, media transcoding, in addition to IO enhancement, such as network enhancement, local disk and other forms. Several combinations, each generation has dozens of hardware architecture forms. Amazon Cloud Technology can roll out new instances quickly, thanks to Amazon Nitro's virtualization offload.

Self-developed Amazon Graviton - high performance and low price

As Intel squeezed toothpaste in recent years, the cost performance of server processors has slowed down, especially the rapid increase in energy consumption. Amazon Cloud Technologies launched the Amazon Graviton 2 and 3 series of ARM processors.

According to the recent analysis of ARM technology, it is speculated that the Nerverous V1 architecture, code-named Zeus, is 50% higher than the N1 core performance (the actual SVE improvement is even greater)

Innovative cloud native chip design concept:

• Performance: Single Core Memory Bandwidth and Latency
• ML: Leading the way in BF16 and INT8
• Power consumption: Power consumption is ARM's strength, and the efficiency is 60% better than x86
• Abandon: As cores increase NUMA as a single-plank bridge, which increasingly affects performance, simply abandon NUMA
• Scenario: Single-core performance is strong, and HPC and media transcoding scenarios have advantages;

Amazon Graviton 3 has 20 billion more transistors than G2; but the processor frequency does not increase, and the number of cores does not change. It is speculated that the strategic positioning is as follows:

• Performance strategy: Choosing the V1 architecture means taking the performance route, focusing on ML and HPC
• Higher cost: Amazon Graviton 3 is only 64core, compared to other processors in the industry, the core density is higher, Ampere AltraMax, Yitian, NVIDIA are 128core and 144core (2chips)
• Benefit customers: Amazon G3 instance C7g pricing is only 5% higher than the previous generation Amazon G2; 5nm 50 billion transistors are expected to double the cost of 7nm 30 billion transistors.

ML chips enable AI scenarios

• 70% lower cost
• Alex has moved to the Inf1 instance
• 35% lower cost
• Throughput increased by 2.3 times

EFA supports the SRD protocol to help customers optimize the parallel capabilities of HPC and ML, and reduce the proportion of TCP communication overhead.

Cloud innovation technology empowers business

AI empowers business

CTO motto: Every line of code is business logic. As can be seen from the OPPO case, the use of Amazon EC2-Inf1 on the cloud to quickly build a voice assistant case.

Intelligent operation and maintenance

Amazon Cloud Technology uses its accumulated experience in e-commerce to help customers operate and maintain through Amazon Cloud Technology cloud services. Including event analysis and machine learning in response. For example, trusted advisor provides customers with more than 50 million intelligent operation and maintenance recommendations. According to the accumulated experience of Amazon's e-commerce, it is output to customers, in addition to tools such as IEM.

data value

Data creation for customers is also very exciting, allowing people who only understand SQL to play with AI. Without going into details, I put the playback link at the end of the article, if you are interested, you can learn more.

Data services provide:

Quick recovery of hot and cold data, data backup and security
Amazon S3-redshift, open up lake and warehouse data security, audit strategy using SQL to achieve ML innovation

Think and inspire

Choose because of value

By saving costs for customers and creating business value, Amazon Cloud Technology is popular with global users. In addition, cloud computing consumption accounts for half of the Internet companies, saying goodbye to involution and going global is the theme of 2022. Globalization requires strong product capabilities with security and compliance features, and a global network. Therefore, domestic companies have chosen Amazon cloud technology to go overseas.

According to the statistical report, although Amazon cloud technology only accounts for 6% in the domestic market, the share of Chinese companies going overseas accounts for 26%, which is enough to show the strong competitiveness and customer stickiness of Amazon cloud technology.

Can you learn the secrets of Amazon cloud technology?

The meaning of the existence of IT service providers is to create value for customers. Finally, let me ask a few questions, is your corporate culture really customer first? This year, revenue KPI and customer stickiness need to be sacrificed. How do you choose?

After cracking the mystery of cloud computing growth, can you learn the product way of Amazon cloud technology?
(End) Codewords are not easy, please like and support as I said earlier, there are many places in this article that have not been explored, and those who love technology like to get to the bottom of it, and attach a playback entry to see it for yourself.