Subscribe

Huawei prioritises new-gen AI data storage for Africa


Johannesburg, 10 Jul 2024
Dr Peter Zhou, president of the data storage product line, Huawei.
Dr Peter Zhou, president of the data storage product line, Huawei.

Huawei Technologies has prioritised the application of data storage solutions and the need for Africa’s ICT sector to pay closer attention to this technology to benefit from digital transformation.

The multinational technology firm recently hosted the Huawei Africa Connect 2024 conference in Johannesburg, during which it announced new product releases across its solution portfolio – with data storage a prominent focus.

Dr Peter Zhou, president of the data storage product line, said: “Applications drive the continuous evolution of data storage. As applications change, data storage continues to evolve from direct-attached storage to storage area network, then to network-attached storage and finally to unstructured storage.”

Huawei contends the rapid development of generative artificial intelligence (GenAI) calls for a new generation of storage in the form of AI storage.

“As AI technologies further develop, a massive amount of data will be generated and made constantly available for use in training large AI models, meaning 'cold' data will turn into 'warm' data, and ultimately, into valuable assets. This process, often referred to as a 'data awakening', heralds a new era where enterprises are becoming data factories, generating an ever-increasing volume of data,” Dr Zhou continued.

“AI can now be employed to mine value from every bit of data, making a robust foundation of data infrastructure more important than ever. It’s time to build an AI-ready data infrastructure capable of accelerating the entire process of AI services, which will entail improving the performance of storage clusters to enhance the availability of computing power.”

Dr Zhou added that while GenAI is booming, it faces challenges that affect its development. “This includes cluster utilisation, which generally falls below 50%. The GPU spends most of its time waiting, and all AI models are power-hungry.”

To this end, Huawei redefines data storage by focusing on ultra performance, new data paradigm, scalability, sustainability and data fabric.

OceanStor A800

To help address these challenges, Huawei has released the OceanStor A800, a powerful addition to the Huawei OceanStor A series storage models.

Dr Zhou said: “Tailored for AI applications, OceanStor A800 can increase AI cluster utilisation by 30%, and as for performance, it delivers high bandwidth and IOPS, which are four and eight times better than its peer vendors. Regarding scalability, OceanStor A800 supports scaling out to EB-level capacity with up to 512 controllers, as well as scaling up to a maximum of 4 096 computing cards.

“As for conserving space and energy, it achieves an outstanding storage density of 1PB/U and energy efficiency of 0.7-watt/TB. It also provides a new data paradigm with vector index, tensor data and RAG. In terms of data resilience, the accuracy of ransomware detection is improved from 99.9% to 99.99%. In addition, the data fabric capability facilitates data asset management.”

Dr Zhou added that GPUs must often wait for a long time to load the data before they start training. They are also idle while reading or writing checkpoints during model training.

“With OceanStor A800, we can significantly improve bandwidth, IOPS and reliability, and shorten the time of dataset loading and checkpoint reads and writes.

This also shortens the GPU idle time, which improves the utilisation of the entire AI cluster by at least 30%.

At the same time, Huawei has released high-capacity SSDs. “Our SSDs deliver 10 times higher capacity over mainstream SSDs at the same size. For 1PB of capacity, our SSDs occupy 88% lower space and consume 92% less power,” said Dr Zhou.

Huawei continues to break new ground in data management, protection, application and impact, it says. Its underlying intention is to always make data assets visible, manageable and available.

Share