Self-practice of high flexibility and high speed AMD cloud computing

At present, cloud computing is not only the hot topic of people, but also has been used more and more. As the world's leading semiconductor manufacturer, AMD has the industry's leading APU, CPU and GPU technology and products to provide a full range of product solutions from cloud servers to cloud terminals. AMD, on the other hand, is the front in cloud computing deployments. AMD has created a private cloud to improve its productivity, improve its ability to acquire resources, save on related costs, and maintain its position as an innovator in the industry. Its self practice in cloud computing is very instructive.

AMD is an innovative traditional semiconductor design company, its various innovative technologies to create a new face of modern computing. With the rise of cloud computing, AMD has once again played the role of technology leader, leading the new generation of energy-efficient multi-core processor technology trends, is the world's cloud computing deployment to lay the cornerstone.

AMD faces complex challenges

The design of semiconductors is a complex engineering task, which requires a great deal of computational power. For AMD, this computing power must be easily and fully accessible to AMD engineers worldwide and multidisciplinary. To this end, AMD has built a number of system clusters to meet computing power requirements, including the deployment of more than 15,000 servers. These servers are equipped with a variety of operating system and file systems, as well as disparate storage management and clustering software. As the annual infrastructure expands 20–30%, hardware and software upgrades will almost never stop.

However, the challenge of "complexity" goes beyond the hardware operating system. Data management and migration are equally problematic. Because the facilities are global and disconnected, AMD has to replicate large datasets between sites. Therefore, it may take several weeks to establish the available project data at a given location.

Once the project is launched, new obstacles will be encountered. Testing the semiconductor design requires acquiring tens of thousands of computing cores. However, in order to avoid waste of resources, any AMD facility will not be overly configured. In order to complete the test and meet AMD's stringent product standards, engineers have to corners time and queue up to run tests on available machines. On the one hand, AMD has more servers than needed, and on the other hand, it is difficult to move enough computing power to a particular project group, even in a very short time. In short, AMD has great computing power, but it cannot meet the requirements of flexibility, manageability, availability, and cost efficiency.

AMD migrates infrastructure to cloud

In April 2009, the AMD IT department started a project designed to change the isolation of servers and storage by consolidating all devices across the globe into a private cloud. The overall strategy of the project – in the words of AMD – is "anywhere." The plan is to centralize the data centers and provide the right connectivity tools to build the AMD private cloud to provide computing power for all AMD design projects worldwide.

Early in the project, the AMD IT department replaced the increasingly obsolete processors in the corporate servers with newer, more efficient AMD Hao long processors. The seamless upgrade of the processor has great benefits for the AMD private cloud. AMD's IT department, while leveraging existing infrastructure, doubles its computing capacity, saving more than $6 million trillion in cost.

At the same time, AMD's IT department has launched another project to standardize software and its versions throughout the new AMD cloud. Based on its multi-core support and high scalability, Rad Hat Enterprise Linux is selected as the basic operating system. After investigating usage patterns and software capabilities, the AMD IT department has also implemented platform computing standardization and optimized design and simulation tools for workload management and EDA software from Cadence, Mentor Graphics, and synopsis.

(Responsible editor: The good of the Legacy)