With every Amazon annual report, Jeff Bezos attaches a copy of the original 1997 letter to shareholders. In the letter, Bezos outlines the basic measures of Amazon's success: a relentless focus on the customer, creating long-term value over the short-term profits of the business, and continuing to innovate boldly. "If we execute well, every day is 'Day one,'" Bezos wrote.
In 2006, Amazon launched Amazon Web Services, which has since created the cloud computing market. For the past eleven years, Amazon has held a strong leadership position in Gartner's Magic Quadrant for Cloud Infrastructure and Platform Services (CIPS). Under the general profit dilemma of the cloud computing industry, in February this year, Amazon announced the data for the fourth quarter of the previous fiscal year. Amazon cloud service revenue this quarter was 17.8 billion US dollars, a year-on-year increase of 40%, and the revenue growth rate will be 4 consecutive years in 2021. continued to increase over the quarter.
Today, the competitors in the cloud computing field are becoming more and more mature, and the market competition is extremely fierce, but Amazon has always maintained its leading position, has been caught up, but has never been surpassed. If you look at the birth and development process of Amazon cloud technology, you will find that the Day one concept of "quick product update and technology iteration based on customer needs" runs through it all the time, allowing Amazon to lead the cloud computing market all the way.
The birth of cloud computing
The technical idea of Amazon cloud technology was actually born 20 years ago, which is very forward-looking. Amazon's Prime service that year brought explosive growth in the number of users, so Amazon began to plan a second project called "Merchant.com" to provide an online experience for large retail customers.
A lot of engineers were put into the project, but it didn't deliver on schedule. Because at the time Amazon.com was a monolithic application with large and multi-platform components intertwined. To add new functionality to customers, developers had to edit and rewrite a lot of code from scratch on this monolithic program, mainly computing and databases. In addition, every time the team solves a problem, it is limited to solving the problem in its own project, and everyone is doing repetitive work.
To speed up the iterative process of the application, Amazon has cleaned up the business and re-architected in a very rigorous way, reorganizing monolithic programs into smaller parts called "services", and dismantling the functional hierarchy of developers , forming numerous small "two pizza teams". Each small team is focused on a specific product, service, or feature set, giving them more authority over specific parts of the application to speed up the decision-making process for the products they are responsible for. In the end, the Merchant.com service was successfully launched, Amazon provided users with an API to set their own prices, titles, and availability, and without any promotion, thousands of developers flocked to this API. And this "microservices" idea that was used to build the Merchant.com application continues to this day and has revolutionized the way modern applications are built.
Breaking down the application and organizational structure is only half the story, the other half is about "infrastructure".
In 2003, Amazon.com's engineering manager Black wrote a short paper on a way to restructure Amazon's infrastructure, raising the possibility of "selling virtual servers as a service." While not related to the retail business, Bezos clearly endorsed and promoted it.
In 2006, Amazon Cloud Technology officially launched their first three products: EC2 (Elastic Computer Cloud), S3 (Simple Storage Service), SQS (Simple Queue Service). Amazon has broken down all IT infrastructure into the smallest units, including networking, storage, computing, and more. Developers are free to choose these units, as well as the software services provided by Amazon Cloud Technology, to build their own products.
Dr. Werner Vogels, Amazon CTO, said, "Amazon wants to use its expertise in scalable system software to deliver raw infrastructure through a service interface, freeing developers from focusing on building and maintaining infrastructure. It's about 'innovation' ' story, the 'cloud computing was invented because of excess capacity' is not true because within 2 months of Amazon Web Services' launch, all of Amazon.com's excess capacity was exhausted."
Over the next decade, Amazon gradually introduced more services, Amazon Simple DB, Amazon CloudWatch, Amazon Route53, Amazon CloudFormation, Amazon Dynamo DB, and more. With the improvement of services, more and more enterprises have moved to the cloud. Slack, a fast-rising unicorn, shared in 2015 how they built it: using Amazon EC2 instances for computing, Amazon S3 for storing user-uploaded files and static assets, and Elastic Load Balancing to balance Amazon EC2 instances between workloads, and nightly backups of MySQL instances running on Amazon EC2 instances using Amazon Elastic Block Store (Amazon EBS).
Source: https://aws.amazon.com/cn/solutions/case-studies/slack
Just like Slack, different enterprises can choose different services according to their own needs and form their own IT architecture. Enterprises no longer need to solve basic problems by themselves and avoid the repetitive work of building wheels. It can be said that cloud technology is a resource optimization method and a systematic innovation method obtained after Amazon's technology has developed to a certain extent.
As of 2021, Amazon Cloud Technology includes more than 245 products and services, including computing, storage, networking, databases, analytics, deployment, management, machine learning, developer tools, and more. Although numerous, this modular, unitary technical solution is particularly harmonious and fluid in design as a whole. According to media reports, some well-known domestic technical experts sighed when they participated in the 2016 re:Invent: "Looking at the architecture of Amazon Web Services, there is a feeling as if the entire Amazon Web Services is done by one person. Modules are defined, reused, and communicated. It was a delight to watch.”
Refactoring the Cloud Base: Building an Evolvable System
calculate
In August 2006, Amazon Elastic Compute Cloud (EC2) opened beta testing and has since launched the cloud computing revolution. Without this innovation in computing power, we take things for granted today—from dispatching calculations in food delivery and express delivery, to genetic computing in life sciences, all of which cannot be easily realized.
Amazon's original idea was to provide developers with on-demand access to computing infrastructure and have them pay only for the resources they use.
Dave Brown, vice president of EC2 services, recalled: "When we first established our R&D center in Cape Town, South Africa, the Amazon cloud technology team was only eight people, but what we were doing was ahead of the times. So much so that when we started in August 2006 When the product was released, most of the comments on Reddit and Slashdot said it was difficult to understand what we were doing. A century ago, many large manufacturers in the United States generated their own electricity to power their factories, and as the grid spread, companies began to shut down their own Power plants, because they have more economical access to electricity anytime, anywhere. The same should be true in computing, like the early grid shift, computing power can be turned on and off when you need it.”
Amazon initially chose to implement the EC2 architecture by modifying and customizing on the open source Xen, virtualizing the CPU, storage and network through the Xen hypervisor, and providing rich management capabilities, allowing multiple virtual machines (VMs) to run on a single physical machine. , with isolation and multi-tenancy between virtual machines provided by the hypervisor.
Xen Hypervisor EC2 Architecture, Source: https://www.allthingsdistributed.com/2020/09/reinventing-virtualization-with-nitro.html
Over the years, EC2 has become a large, robust environment and a multi-billion dollar business. While traditional virtualization architectures have been optimized to the limit by Amazon, with this architecture up to 30% of the resources in an instance are allocated to the hypervisor and monitoring operations for network and storage. To reduce these losses, Amazon Cloud Technologies began customizing dedicated hardware, reinventing the underlying EC2 technology from the ground up.
This didn't happen overnight, and in 2013 Amazon released EC2 C3 instances that offload networking processes to hardware (moving functionality from software to hardware). In 2014, Amazon Cloud Technology partnered with Annapurna Labs to launch EC2 C4 instances again. The C4 offloads the EBS storage to the hardware, but still relies on the Intel Xeon chip as the controller. Amazon Cloud Technology reportedly spent $350 million to acquire Annapurna in 2015, introduced C5 instances, replaced Xen with the Nitro hypervisor, and tightly coupled virtual machines to ASICs. Werner Vogels said the milestone offloaded the control plane and the rest of the I/O, used nearly 100 percent of the processing to support customer workloads, and also enabled a bare-metal version of compute, spawning a partnership with VMware to launch VMware Cloud on AWS. In 2017, Amazon officially launched the full version of Nitro.
Nitro is a set of custom hardware and software designed to offload hypervisor, networking, and storage virtualization to dedicated hardware, freeing up CPUs to run more efficiently.
Nitro system architecture in 2017, source: https://www.allthingsdistributed.com/2020/09/reinventing-virtualization-with-nitro.html
In 2018, Amazon Cloud Technology launched Graviton, an Arm-based custom chip. According to relevant data, the price/performance ratio of Arm-based Graviton2 is 40% higher than that of similar x86-based instances. This broke the reliance on x86 and ushered in a new era of architecture, now Amazon has been able to support intensive computing tasks in a variety of configurations. These moves lay the foundation for other innovations at Amazon Cloud Technology, including optimizations for machine learning and artificial intelligence tasks from training to inference.
Since the introduction of the Nitro System, EC2 instances have also increased rapidly, and Amazon EC2 now has over 475 instances. Computing methods are also constantly being updated, starting with EC2 instances and gradually supporting containers and serverless. General cloud platforms only provide Kubernetes hosting. Amazon Cloud Technology provides three container management tools: EKS (Managed Kubernetes Service), ECS (self-developed) and Fargate. Amazon Lambda also ushered in the era of serverless computing, a way to provide backend services on demand. Serverless providers allow users to write and deploy code without worrying about the underlying infrastructure.
storage
In 2006, Amazon Cloud Technology launched the S3 (Simple Storage Service) service. S3 defines object storage and is the de facto standard for object storage, with epoch-making significance.
The launch time of S3 was actually 6 months earlier than EC2. An important principle of the initial design was "simple", so only GetObject and PutObject functions were provided at that time, and the core API only had four REST commands (GET, PUT, LIST and DELETE). Werner Vogels and Tom Killalea, the former VP of technology at Amazon, discussed the development of S3, and the two believe that "'as simple as possible' is one of the keys to S3's success, and although this was controversial at the time, an evolvable architecture-- It couldn't have been designed to be complicated to begin with."
In the beginning, users mainly used S3 to store image and video data, but over time, more and more transaction logs, parquet files, customer service records and other data were put into S3. Then users didn't just want "elastic storage" and "low cost", they wanted to be able to use those shared datasets for hundreds of applications, so Amazon added the "Access Points" feature. This process is enough to illustrate the evolutionary principle of S3: what users want to do with their data, Amazon cloud technology adds any function or service. Through technical and commercial services, Amazon Cloud Technology has established a rapid feedback loop with users and has become a fast-growing flywheel.
In 2006, S3 had only 8 services (Services), and by 2019, S3 has 262. And Amazon is committed to making S3 higher performance and lower cost, and has launched seven different levels of storage products: Standard, Intelligent Tiering, Infrequent Access, One-Zone Infrequent Access, Glacier, Glacier Deep Archive, and S3 Intelligent Tiering. The S3 Intelligent Tiering product is further divided into two tiers: "frequent access" and "infrequent access", which will automatically move objects that have not been accessed for 30 consecutive days to the "infrequent access" tier, reducing operational complexity . In 2021, intelligent tiering will also be increased from 2 to 3 tiers, a newly launched archive instant access tier will be added, and the most classic storage will continue to evolve.
S3 storage itself has the characteristics of separation of computing and storage. In the cloud-native era, it is very suitable as the core of data lake storage. Enterprises can build data lakes based on Amazon S3, while leveraging native Amazon Web Services services to run artificial intelligence or machine learning services (SageMaker), which can process various structured and unstructured data more efficiently.
S3 durability is designed to be "11 nines" (99.999999999%), meaning that using Amazon S3 to store 10,000,000 objects, you can expect an average of one object loss every 10,000 years. This year, Amazon announced that the number of objects stored on S3 has exceeded 200 trillion, and it needs to process tens of millions of requests per second. Today, S3 has evolved into a large and robust distributed storage system. In order to maintain data persistence, Amazon announced at the end of last year that it had upgraded S3's storage backend system, ShardStore, and introduced an "automatic reasoning" method to ensure "crash consistency". ”, that is, the data can still maintain a consistent state of “11 9s” when the system crashes.
The implementation of ShardStore is relatively complex, including more than 40,000 lines of rust code. It uses soft update to provide crash consistency. The speed of traditional verification methods cannot keep up with the speed of system development iterations. Amazon adopts lightweight formal methods to improve the reliability of ShardStore, automatically generates a series of operations to call the interface of the key-value storage system, and checks the behavior and relationship between the reference model and the specific implementation of ShardStore during the operation. Whether the status is consistent. Based on the above method, Amazon successfully found and fixed 16 important issues in ShardStore, involving bugs such as crash consistency and concurrency. Amazon Cloud Technology also published a related paper at SOSP and won the Best Paper Award.
network
The network is one of the most basic supports for cloud computing business. Amazon Cloud Technology has the largest global backbone network, with 25 regions, 81 availability zones, 14 Local Zones, 17 Wavelength Zones, and more than 300 edge sites. and 108 dedicated line access points. These are based on the single flat subnet in 2006, after 16 years of continuous innovation and optimization.
Under the wave of globalization, many large multinational companies will set up multiple sites around the world. When building a global network, users of Amazon Cloud Technology can use Amazon VPC to create multiple virtual networks. The traditional approach is to use the VPC Peering function to connect VPCs within or across regions, and use Direct Connect or VPN to interconnect non-Amazon cloud technology infrastructure with Amazon cloud technology. However, if the load on the cloud increases, the management work will increase exponentially. At this time, a centralized management link solution can be adopted. Amazon launched the Amazon Transit Gateway at Re:Invent in 2018. With Amazon Transit Gateway, management is significantly simplified and operational costs are reduced because each network only needs to connect to the Transit Gateway, rather than to all other networks.
As networks extend across the globe and also require a mix of technologies, the complexity of building, managing and monitoring them increases significantly. At re:Invent 2021, Amazon Cloud Technologies announced the Cloud WAN global web hosting service. With this network service, enterprises can use Amazon's backbone network to create their own global network with one click using the Cloud WAN graphical interface, set up transit gateways or cloud connections, and monitor network health, security, and performance.
Source: https://aws.amazon.com/cloud-wan/
In a keynote and blog post at re:Invent in 2021, Werner Vogels also mentioned Amazon's vision for "the everywhere cloud", one that connects Amazon with targeted hardware and solutions. Bringing cloud technology to new territories.
Werner Vogels wrote in a blog post: "We will see a shift in 2022: the cloud will become highly specialized at the edge. We will provide tailor-made solutions for the edge, whether it is a workshop, restaurant, small retail store Or remote areas, you can take full advantage of the cloud.”
Safety
Without the guarantee of security, all the advantages of the cloud will be impossible to talk about. As millions of organizations migrate to the cloud, data and traffic are more concentrated, and the cloud has long become the main battlefield for security attack and defense.
As a pioneer of cloud computing, the "Security Responsibility Sharing Model" pioneered by Amazon has become a de facto industry standard recognized by the Cloud Security Alliance. This model defines the security boundary between cloud vendors and tenants, and also clarifies the internal Safety Responsibility. This year, Amazon Cloud Technology once again proposed five layers of "onion protection methodology": threat detection and incident response, identity authentication and access control, network and infrastructure security, data protection and privacy, risk management and compliance. Based on more than 280 security and compliance services and functions, Amazon Cloud Technology uses these five-layer protection systems to provide customers with a full range of security services.
sustainable development
The United Nations formulated a global framework, the Paris Agreement, in 2015. Subsequently, the signatories have formulated “carbon neutrality” paths and goals, making commitments to the healthy development of the global environment. According to Accenture analysis, a green approach to moving to the public cloud could reduce global carbon dioxide emissions by 59 million tons per year, which is the equivalent of cutting carbon emissions from 22 million cars at your fingertips!
As a world-class technology giant, Amazon has led the transformation of the "green cloud". Amazon has said it will reach the Paris Agreement ten years early and achieve 100% renewable energy by 2025, and has designed a forward-looking solution from infrastructure to software design. These initiatives have also yielded notable results. In 2021, Adrian Cockcroft, vice president of sustainable development architecture at Amazon Cloud Technology, said: Amazon Cloud Technology's infrastructure energy efficiency is 3.6 times higher than that of the average US enterprise data center. At the same time, Amazon can reduce its carbon footprint by 88% when performing the same task.
Reimagining the Cloud Base: Innovating for Customers and the Future
Amazon Cloud Technology, which was born 16 years ago, created a new field of cloud computing. Amazon's innovation can be said to have had a revolutionary impact on the evolution of the IT industry.
Although the competition among cloud vendors is fierce today, Amazon Cloud Technology has always been in a leading position in the market. According to relevant data, Amazon Cloud Technology has the largest share of cloud infrastructure service providers at 33%, and customers have already exceeded one million. Whether it is technology giants, banks or governments, different organizations are using Amazon Web Services to Develop and deploy your own applications.
Early typical users are Netflix, which has adopted Amazon Web Services since 2009, and Netflix closed its last data center in mid-2015. Nasdaq, which has been using Amazon Web Services to store stock exchange data in the cloud since 2014, increased its use of edge solutions again this year, gradually migrating Markets to Amazon's cloud services. NASA began in 2000, using Amazon Web Services to provide online access to photos, video and audio, and last week, NASA once again announced the use of Amazon's cloud services to build a "space logistics" infrastructure system for the space station...
Such achievements are attributed to Amazon's continuous technological innovation around customer business. Some industry experts believe that a major highlight of Amazon's cloud technology is that it can very sensitively discover what problems users are currently facing urgently, and quickly provide solutions or products. It's also like what Dave Brown said: "We can handle almost any need and challenge and never say no to our customers. At Amazon Cloud, we have a 'customer-first' culture that not only meets their current needs, but also They will predict their future needs.” Customer-centric, continuous innovation, which is also a manifestation of the “Day one” philosophy.
In the process of future-oriented development, Amazon Cloud Technology undoubtedly has many technical concepts and decision-making experience in the past 16 years that we can learn from and think about.
On April 20, 2022, Amazon Cloud Technology will hold the INNOVATE innovation conference online, which is also Amazon's first technical conference dedicated to cloud infrastructure. It will set up 6 sub-venues and 30+ cutting-edge technology topics, which will be shared more comprehensively and in more detail. The underlying innovation of Amazon cloud technology, interpreting the application of different business scenarios, and helping you build a data-driven enterprise!
Reference link:
- https://www.allthingsdistributed.com/2020/09/reinventing-virtualization-with-nitro.html
- http://muratbuffalo.blogspot.com/2021/10/using-lightweight-formal-methods-to.html
- https://www.youtube.com/watch?v=6ryVoGlgwU4
- https://perspectives.mvdirona.com/2016/03/a-decade-of-innovation/
- https://thestack.technology/aws-shardstore-s3/
- https://dl.acm.org/doi/fullHtml/10.1145/3434571.3434573
- https://www.allthingsdistributed.com/2021/12/tech-prediction-for-2022-and-beyond.html
- https://aws.amazon.com/cn/blogs/publicsector/amazon-aws-reimagine-space-station-operations-logistics-orbital-reef/
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。