Author: Hao Junhui Source: IT Times
Thirteen years ago, when " Avatar " conquered the world with its gorgeous "Pandora Planet Scenery" for the first time, people didn't know that director Cameron built a huge offline data center with 4,000 servers and used more than 35,000 CPUs to render the entire movie, with a rendering time of 2.88 million hours.
Today, 13 years later, the rendering time of the azure water world in the currently being screened "Avatar 2" reached 3.3 billion hours, which is 1145 times that of the first film. However, it only took 8 months from the first frame to the 3.3 billionth frame. "Without the support of Amazon Cloud Technology, we would not be able to complete Avatar 2." The producer of "Avatar 2" said.
The difference in results between creating a virtual world and a simulated world is obvious whether or not the cloud is used, and this is just one of the cases where Amazon cloud technology is applied globally.
At the 2022 re:Invent global conference held not long ago, Amazon Cloud Technology released a series of blockbuster new services, products, functions and applications, covering underlying infrastructure, computing, databases, data analysis, chips, AI/ML, security, industry applications, etc. Many of the cases were announced for the first time, which can be called Amazon Cloud Technology’s “annual collection of useful information”.
For the Chinese government and enterprises that are accelerating the digital transformation of to , these innovations and reshaping at the forefront of technological trends are of great reference significance.
Nasdaq : Building the next generation cloud infrastructure for the securities industry
Last year at re:Invent, Nasdaq announced that it would move its core transaction processing engine to Amazon Cloud Technology. At this year's re:Invent, Nasdaq "Call Back" this decision, announcing that its MRX core trading system has been moved to Amazon Cloud Technology.
Trade matching is a core system that converts orders into transactions. It is the cornerstone of the infrastructure of the global capital market, and MRX is one of the six U.S. options markets owned by Nasdaq. These workloads mean that they are not only highly sensitive to overall latency, but also highly sensitive to security. The completion of the migration marks Nasdaq's construction of a next-generation infrastructure in the global capital market.
Amazon Outposts plays a central role in Nasdaq's migration to Amazon Cloud Technology. It extends Amazon Cloud Technologies' infrastructure, services, APIs and tools to virtually any data center, colocation space or on-premises facility, allowing Nasdaq to deliver ultra-low latency edge computing capabilities from its primary data center in Carteret, New Jersey. These new cloud systems built on Amazon Outposts can provide a double-digit microsecond latency for end-to-end, order-to-transaction, and achieve a 10% performance improvement in two-way transaction latency.
That's not all. Nasdaq currently has more than 130 market technology customers, including exchanges, banks, clearing houses, central securities depositories and regulatory agencies, which currently rely on Nasdaq's core trading, clearing, settlement and monitoring technologies. In the future, Nasdaq plans to cooperate with Amazon Cloud Technology to develop a variety of feasible cloud service models, including public cloud and hybrid cloud, to provide cloud migration services for its market technology customers.
Netflix: Maintaining high-performance analysis of massive data
In 2012, Netflix participated in the first re:Invent. As the world's leading integrated media provider, Netflix transformed into streaming media in 2006 and became a customer of Amazon Cloud Technology in 2008. Today, traces of Amazon Cloud can be seen in every aspect of Netflix.
For example, the fourth season of the hit drama "The Crown" used a cloud studio based on Amazon cloud technology to solve the production problems caused by the global epidemic of the new crown epidemic. The internal creative team of "The Crown" composed of 10 artists collaborated remotely to successfully complete more than 600 shots in the 10 episodes of the fourth season in only eight months, and the production cost was also greatly reduced.
Another important direction for Netflix to use Amazon Cloud is how to cope with the arrival of the era of massive data. Chen Xiaojian, general manager of Amazon Cloud Technology Greater China Product Department, gave a set of data: Expedia is the world's leading travel service provider, with more than 70 PB data, and makes 600 billion AI predictions every year; Samsung has more than 1.1 billion end customers and receives 80,000 requests per second; Pinterest is a service provider that provides video and picture websites in North America. The image data they store on Amazon Cloud Technology exceeds 1 EB, which is equivalent to 1 million TB, and What about Netflix? It must serve one billion pieces of user traffic every day. At the same time, in order to support the best user experience, nearly TB of log data must be collected and analyzed from all production systems every day.
The data generated in the next five years will be more than double the data generated before. The data will be as vast as the starry sky. It is no longer the future, but reality. How to manage massive data and mine the value of data is a huge challenge for every enterprise on the one hand, and it also contains unlimited value on the other.
What these Internet giants need to think about is how to maintain high-performance processing while data is growing rapidly? This re:Invent also gives the answer.
Amazon Aurora is Amazon Cloud Technology's cloud-native relational database service. Its automatic expansion function can automatically expand each database instance to a maximum of 128 TB. The usage cost is about one-tenth of that of traditional commercial databases. Amazon DynamoDB is the industry's first cloud-native database. It will handle more than 100 million requests per second on Amazon Cloud Technology Member Day in 2022. It is a very high-performance interval database. Amazon Redshift has a large number of users, and the amount of Amazon Redshift data processed every day exceeds one EB. The price/performance ratio is five times that of other cloud data warehouses.
In addition, Amazon DocumentDB's Elastic Clusters can not only support millions of read and write requests per second, but can also expand the storage capacity of Amazon DocumentDB from 64TB to two PB by 32 times.
Avatar: A high-performance computing ecosystem composed of chips, protocols, etc.
As mentioned above, "Avatar 2" uses Amazon cloud technology services, which greatly shortens the time for visual effects production. Creating a virtual world based on 3D, like the concept of "simulation" and "digital twin" that became popular last year, requires a lot of complex high-performance computing.
Amazon Cloud Technology has won the title of HPC WIRE's Best HPC Cloud Platform for seven consecutive years. This should be due to the complete HPC service ecosystem it has built over many years.
The most important thing in the ecosystem is naturally the chip . Through 10 years of research and development, Amazon Cloud Technology has built three product lines of self-developed chips, including the fourth generation of customized chip Nitro, the third generation of ARM architecture-based CPU processor Graviton, and two chips for machine learning training and inference.
Chen Xiaojian introduced that Graviton3E, for example, is the latest chip specially designed for high-performance computing. It is equipped with 200Gbps low-latency bandwidth and is very suitable for computing-intensive HPC applications. It is optimized for floating-point and vector instruction operations, such as computational fluid dynamics, molecular dynamics, weather simulation, etc. From the perspective of user scenarios, in the new generation of HPC data measurement tools, Graviton 3E has a 35% improvement compared to Graviton 3. In the molecular motion analysis scenario, the performance can be improved by 12%, and in the financial options pricing, the performance can be improved by 30%.
In the entire self-developed chip family of Amazon Cloud Technology, in addition to Graviton, there is also Nitro. Nitro is a chip that professionally hardware-based the underlying architecture and virtualizes the cloud server . It has now developed into the fifth generation. From the 10Gbps network performance of the first generation Nitro to the current 100GB, the performance has increased by 10 times. The number of Nitro V5 transistors is twice that of the previous generation Nitro, the forwarding capacity of the entire data packet is increased by 60%, the delay is reduced by 30%, and the performance of per 1 watt per html is increased by 40%.
In addition, the network protocol SRD developed by Amazon Cloud Technology is also a "unique weapon" to support high-performance computing . Chen Xiaojian gave the video stream as an example. Using SRD, the video stream can be broken up during transmission, multiple paths can be transmitted in parallel, and the receiver can reorganize it out of order through the sequence number. Through the SRD protocol, the maximum traffic of a single stream on the network has been increased from 5GB to 25GB, an increase of five times. And because the SRD protocol is implemented on the entire underlying Nitro chip of the network, it is completely transparent to users. Users still use the TCP/IP protocol. Nitro will automatically adapt the TCP/IP protocol to SRD, and then restore this to a TCP/IP package after reaching the opposite end.
"You don't need thousands of troops to cross the single-plank bridge, but it can bring extremely high network throughput, extremely low latency, and higher reliability. If there is a problem with any path, it will not affect the interruption of your entire traffic." Chen Xiaojian said.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-