With the vigorous development of the digital economy, data is reshaping the mode of social operation, promoting the innovation and development of all industries, and becoming a key driving force for economic development. Nowadays, data has become an important asset of enterprises. Enterprises hope to use data to understand market and customer needs to make quick decisions. This not only requires efficient data flow, but also has higher and higher requirements for data storage. Traditional storage cannot be expanded efficiently and flexibly, and it is difficult to meet the demands of modern applications of enterprises. Enterprises urgently need new smart storage products.
1. In the new era of AI, new storage challenges
Since the beginning of this year, AIGC has triggered a new round of technological changes around the world, and the storage field is no exception. Guan Tianshu, general manager and chief product manager of the storage product line of H3C Group, made an in-depth analysis. From the perspective of the development of large models, it is currently in the stage of training and generative AI, mainly text data storage. However, in the next round of development, there will be a great demand for storage. After generative AI enters the multi-modal field, the demand for storage will increase sharply, which will bring the following challenges:
One is that in addition to CPU computing, large model applications are more GPU computing, which requires storage to adapt to the acceleration of GPU storage;
Second, the storage training time determines the production time of the large model, so there is a greater demand for storage performance;
Third, the performance of storage products is strongly related to applications, and the requirements for comprehensive storage capabilities under the AI model will be higher;
The fourth is the core. There is a technology called CheckPoint in large model training. It is very important for large model training. It needs to be stored to provide a very stable state in order to restore the previous training to the state before CheckPoint.
To sum up, in the era of large-scale models, there are higher requirements for storage: faster speed, stronger reliability, simpler minimalist power, and sustainable capabilities.
Therefore, on August 22, H3C Group, a subsidiary of Tsinghua Unigroup, released a new generation of smart storage products, H3C/HPE Alletra MP global decoupling NVMe smart all-flash storage, H3C UniStor X10000 G6 series distributed converged storage products, built on the basis of a new technical architecture Powerful performance, ultimate experience, green and low-carbon smart storage new products.
Xu Runan, vice president of H3C Group and president of the computing and storage product line, said that the outbreak of AIGC technology has brought industrial intelligence and intelligent industrialization into a period of rapid development. The iteration of IT infrastructure such as computing and storage must keep up with market demand, so that customers can Focus on upper-level business innovation without worrying about the support of the underlying infrastructure. As a leader in digital solutions, New H3C has always adhered to the IT strategy of "Endogenous Intelligence· Achievement of Wisdom", continuously promotes the intelligent evolution of products, and injects new development momentum into the AI era.
2. Restructure architecture, experience, and value to create a new generation of smart storage products
Artificial intelligence technology represented by large models not only brings new opportunities to storage, but also presents new challenges. As Li Bin, the centralized storage product manager of H3C storage product line analyzed, different from traditional storage, new storage should meet the following requirements: First, storage products should weaken their own existence and not increase the cost and complexity of customers. Get a ubiquitous cloud experience through simplified management, and quickly and easily deploy more application loads and services at the edge and locally. These workloads and data need to be deposited on the same platform to provide the same management experience. Second, in the data-centric era, storage must keep pace with the development of cutting-edge applications and data, and provide efficient resource utilization according to changing needs, so that it can meet users' individual needs for capacity and data. In the face of the rapid development of AI technology, storage needs to be reconstructed, further evolved, and requires new development concepts.
In this regard, New H3C has created a new generation of smart storage products by reconstructing architecture, experience and value to help enterprises accelerate their development in the AI era.
Architecture reconstruction: How to make storage products better adapt to the AI era, release performance, and make storage more minimal? Alletra MP is the industry's first all-flash storage with global decoupling architecture for key business. The control node and capacity node are completely independent to meet the individual needs of enterprises. The X18000 G6 distributed storage released this time integrates 4 distributed all-flash nodes in a 2U space, and is interconnected through high-speed backplane switching to achieve extremely high performance density.
Zhang Ying, product director of H3C Group's storage product line, specifically introduced that Alletra MP has centralized storage with distributed characteristics. The advantages of centralized storage are stability, reliability, and wide compatibility, but when the computing power or node capacity is not enough, it will be limited by software and architecture. And Alletra MP has been greatly upgraded. It has centralized storage with distributed and cloud features, insists on the compatibility of centralized storage, and has distributed features at the same time.
X18000 G6 is a distributed storage product with centralized features. It puts four nodes in one chassis, adds the features of centralized storage, and can achieve fast switching and fast fault recovery. This is a traditional distributed storage product. Storage does not have a function.
Reconstruction experience: Based on the underlying innovation of architecture reconstruction, Alletra MP and the new generation of X10000 will bring customers four major storage experience upgrades: ultimate performance, ultimate reliability, minimalist management, and green and low-carbon storage, making smart storage the key to digital transformation of enterprises new engine.
Taking green and low-carbon as an example, under the double-carbon goal, building a green data center has become an inevitable requirement for the high-quality development of the digital economy. As one of the infrastructures of the data center, storage faces the sustainable development demands of green and low-carbon.
Guan Tianshu said that fundamentally speaking, green energy saving not only saves energy for the society, but also benefits environmental protection, and can also create more value for customers.
Facing green needs, H3C UniStor X18000 G6 distributed high-density flash memory storage, four-node high-density deployment in one frame, increases storage density by 71%, saves cabinet space by 75%, and reduces overall power consumption by 26%. Alletra MP helps reduce the overall PUE through multiple green energy-saving technologies such as CPU and fan and other hardware energy-saving design, data deduplication and compression, and intelligent power management.
Reconstructing value: Efficient, reliable, minimalist, and green storage solutions meet the needs of large-scale data storage and fast access in the AI era, support differentiated storage needs in various segments, and help customers deeply mine data value and release innovation vitality .
But in the end, Guan Tianshu made a fool of himself: this new product release is just the beginning, and there is still a big plan to restructure the storage system in the future. We expect that H3C will bring more innovative storage products and solutions in the future to meet the needs of various industries in the AI era.