Editor | Song Hui
Produced | CSDN Cloud Computing
Amazon Cloud Technology's top conference in China every year - the 2023 Amazon Cloud Technology China Summit has come to a successful conclusion. This year's summit focuses on issues such as AIGC and globalization, and these products and services are supported by Amazon Cloud Technology for more than ten years Since then, we have explored, innovated and accumulated technical strength. On the second day of the summit, Matt Wood, Vice President of Global Products of Amazon Cloud Technology, and Chen Xiaojian, General Manager of Product Department of Amazon Cloud Technology Greater China, introduced in detail Amazon Cloud Technology's core technology analysis and experience in product services for AIGC. For the three aspects of computing power, AI, and data, developers should focus on understanding.
Self-developed chips, highly reliable cloud infrastructure, supporting AIGC's computing power needs
Providing cloud infrastructure and computing and storage resource services is the housekeeping skill of Amazon Cloud Technology. After 17 years of accumulation, Amazon Cloud Technology at the summit listed in detail how to build a highly reliable cloud infrastructure design idea, which is divided into eight aspects :
- Regional isolation, multi-availability zone design : each region will have multiple availability zones, and the failure of any single availability zone will not affect the business of other availability zones, minimizing service interruption.
- Decoupling of the control plane and the data plane : Relying more on the data plane to keep services running and reducing dependence on the control plane can improve overall stability.
- Cellular architecture : Divide the system into multiple small, single, and isolated honeycomb units, which can reduce the explosion radius when a problem occurs, and the fault can be controlled within one unit through the explosion radius.
- Random sharding : It is a further optimization of the cellular architecture. It randomly assigns customer access to different units, saying that the failure of a single unit will not affect the entire system, or it can be completed through other units.
- Service Responsibility Model : By clearly defining the responsibility range between Amazon cloud technology and customers, customers are guaranteed to have control over the written code and programs at any stage of the application.
- Operational readiness review : Through the analysis of a large number of Amazon cloud technology operation cases, the past problems are reviewed, so that users can avoid reproducing their previous problems during the entire deployment.
- Secure Continuous Deployment : Minimizes impact on production due to wrong deployments.
- COE error correction process : understand the system status when the problem occurs, so as to prevent similar errors from happening again.
In addition to the design idea of the technical architecture, another important technical direction, which is also the focus of various cloud vendors, is the research and development of their own chips. In addition to supporting mainstream chips such as Intel, AMD, and Nvidia, Amazon Cloud Technology has also made outstanding investments and achievements in self-developed chips in recent years. Now, Amazon Cloud Technology has launched the fifth-generation Nitro, the Arm-based general-purpose processor Graviton 3, the machine learning training chip Trainium, and the machine learning reasoning chip Inferentia, covering important computing power requirements and scenarios, and in terms of resource consumption and isolation interference. Breakthroughs have been achieved in various technologies, such as extreme performance and encryption.
AIGC R&D tools: large model service Amazon Bedrock, self-owned model library Amazon Titan, free AI code programming assistant Amazon CodeWhisperer
AIGC is an important topic of this year's summit, and we can also see Amazon Cloud Technology's full investment in AIGC. In addition to the above-mentioned reasoning chip Inferentia and training chip Trainium, in the technical keynote speech on the second day, Matt Wood, vice president of global products of Amazon Cloud Technology, introduced in detail that on top of the general large model, for enterprise users, unlock AIGC Four aspects of technical work need to be done for value: first, to provide access to first-class basic models, second, to provide a secure and private environment to customize models, third, to provide low-cost and low-latency access through custom chips; fourth, to provide Search for opportunities to improve user experience.
In addition to Amazon Cloud Technology's well-known machine learning solution Amazon SageMaker, this year's summit also focused on three other products and services for AIGC, namely the large model service Amazon Bedrock, its own model library Amazon Titan, and free AI code The programming assistant Amazon CodeWhisperer improves the efficiency of AI research and development and lowers the development threshold.
The first is Amazon Bedrock released by Amazon Cloud Technology in April this year, which allows developers to use a variety of basic large models such as AI21 Labs, Anthropic, Stability AI and Amazon’s Titan in the form of APIs, Amazon Bedrock and Amazon Cloud Technology Data Lake Tightly integrated with data services, the basic large model and private data can be combined to develop customized models without requiring a large amount of labeled data. In addition, users can use Amazon SageMaker JumpStart to discover and deploy more open source models.
After accessing the basic large model, developers can use the Amazon Titan model library to achieve safe and private model tuning. Amazon Titan is a library of a series of different models, which can implement text summarization, search result embedding, harmful content deletion, etc. Users can optimize and fine-tune these models in a very safe and private manner, and finally realize their own industry and scene models Custom Development.
Another important service is Amazon CodeWhisperer, a free AI code assistant for individual developers. Amazon CodeWhisperer provides developers with a code generation service based on machine learning and supports 15 different languages including Java, JavaScript, and Python. Programming language. In addition to learning from billions of lines of public code, Amazon CodeWhisperer is also based on Amazon's code training to generate the most accurate, fastest, and safest code for cloud services such as Amazon EC2, Amazon Lambda, and Amazon S3. Developers using Amazon CodeWhisperer complete tasks an average of 57% faster and with a 27% higher success rate.
Build the AIGC data base and launch Zero-ETL's cloud-native data platform
In addition to basic computing power and AIGC tools, another key technical direction of Amazon Cloud Technology in recent years is to build a complete one-stop cloud-native data platform, and has gradually demonstrated its strength and advantages. At present, Amazon Cloud Technology has launched 15 cloud-hosted database services to provide data services for various user application scenarios, among which analysis services have fully realized Serverless, such as interactive query service Amazon Athena, big data processing service Amazon Managed Streaming for Apache Kafka (Amazon MSK), real-time analysis service Amazon Kinesis, data warehouse service Amazon Redshift, data integration service Amazon Glue, business intelligence service Amazon QuickSight, and operational analysis service Amazon OpenSearch Service.
With the accumulation of so many data products, at the summit, Amazon Cloud Technology proposed an important data strategy and vision of Zero-ETL, which is committed to realizing seamless data conversion and calling without users needing to write any code . This vision is based on a new service from Amazon Cloud Technologies: Amazon Aurora's Zero-ETL integration with Amazon Redshift , allowing near real-time analytics and machine learning (ML) on petabytes of transactional data from Aurora using Amazon Redshift. Transactional data is available in Amazon Redshift within seconds of being written to Aurora, and developers don't have to build and maintain complex data pipelines to perform extract, transform, and load (ETL) operations.
In addition, Amazon Cloud Technology hopes to build end-to-end data governance to accelerate and ensure data circulation. Amazon DataZone, a new data management service launched by Amazon Cloud Technology last year, allows users to catalog, discover, share and manage data stored on Amazon Cloud Technology, customers' local and third-party sources faster and easier. With Amazon DataZone, administrators and data asset managers can use granular control tools to manage and govern data access permissions, ensuring that data access occurs with the correct permissions and in the correct context.
Summarize
Using computing power and data to support AIGC and innovation on the cloud is easy to say, but actually requires a lot of technology and R&D energy. All the technologies and product services mentioned at the 2023 Amazon Cloud Technology China Summit ultimately hope to provide developers and users on the cloud with "all-round, no dead ends" services and support, which is also the hope of a global cloud computing giant. The goal is to use all the technical products and services of Amazon Cloud Technology in 99 availability zones (covering 245 countries and regions) in 31 regions around the world to provide users with any application, any running scenario, any contract It can provide stable support and services according to the specification requirements and any flexible computing resources. CSDN will continue to report important technical progress of Amazon cloud technology, and it is also recommended that developers focus on it.