Kathy Gibson reports – In the artificial intelligence (AI) era, data storage is gaining new prominence.
Dr Peter Zhou, president of Data Storage Product Line, Huawei, explains that storage has evolved over the years as we use data differently – and in the era of AI, we need AI storage.
“AI is a different application,” he points out. “Because it is data-driven, storage has an important role to play.
“We believe it is the right time to define a new generation of data storage.”
Increasing the performance of the AI data centre is often believed to centre around the addition of GPUs (graphic processing units). Zhou says. But this may not be the most efficient solution as even the fastest GPUs have to wait for data, while using more energy.
Energy is a critical consideration, with studies indicating that AI is already responsible for 50% of the data centre energy consumption. Indeed, by 2026, it is thought that data centre energy consumption will be equivalent to the entire country of Japan, Zhou says.
“So there is an opportunity for us to optimise data storage.”
Traditional storage focused on performance, data paradigms and reliability.
“With AI, we believe that new factors need to be considered,” Zhou says.
These new factors defining the efficacy of storage are:
* Performance – which should be up to 10-times higher;
* Data fabric – which needs to not only store data but help users manage their data assets;
* The new data paradigm – catering to vector, tensor, RAG and more;
* Scalability – not only scale out to accommodate more data, but scale up;
* Data resilience – storage needs to be reliable, with ransomware detection and recoverability built in; and
* Sustainably – there is a pressing need for better energy usage.
Huawei’s answer is the OceanStor A800, announced at Huawei Africa Connect today.
Zhou explains that it offers data control separation, detection flow, embedded ransomware detection, a data fabric, multiple data paradigms, and higher disk capacity.
Performance is 10-times higher in both bandwidth and IOPS; it scales out to 512 controllers and up to 4 096 cards; offers sustainability with 1PB/u running at just 0,7 watt per Tb; handles all data paradigms, and has great data resilience.
AI consumes massive amounts of data, so we need to increase our storage capacity, Zhou points out. To this end, storage media is key.
Huawei has launched a new SSD designed for data centre energy efficiency. It can support 10-times the capacity in the same size packaging, “So we use less space and lower energy consumption can be achieved,” Zhou says.
Increased storage capacity on its own won’t solve the challenges associated with higher data usage in AI applications – being able to manage and identify data assets is crucial.
Huawei today launches new software and data fabric for AI, the Omni-Dataverse.
“This helps users to understand where the data is, have visibility of the asset, gain mobility, and use data efficiently,” Zhou says.