Age of AI - large models call for better database choice

AI, Pangu Models, automation pervade across Huawei’s cloud database, infrastructure, services

Partner Content Huawei Cloud hosted a series of events during MWC24 Barcelona to "Accelerate Intelligence with Everything as a Service", including the Huawei Cloud Summit, Product & Solution Launch, and the Cloud Native Elite Club (CNEC) Seminar.

These occasions were designed to showcase Huawei's infrastructure as a seedbed of artificial intelligence (AI) innovations, with executives and experts from diverse industries mulling over the holistic impact as well as the game-changing possibilities of blending AI with databases, big data, media services and computing architecture.

Huawei Cloud presented GaussDB at all three events. During the Cloud Summit, Bruno Zhang, CTO of Huawei Cloud, introduced the GaussDB distributed relational database, which is commonly used in the finance, telecom and government sectors. "A strong database is a solid data foundation that supports intelligent digital upgrades," emphasized Mr. Zhang.

By applying AI capabilities to database migration, deployment and maintenance, GaussDB makes data processing more efficient for AI applications, said Huawei.

The latest iteration of the database, which was introduced in 2023, boasts round-the-clock service availability, consistent high performance under heavy loads, and stability during high concurrency scenarios. It also couples CC EAL4+ certification with secure source code and end-to-end encryption measures.

Dr. Nikos Ntarmos, Director of Huawei Central Software Institute's Database Lab, emphasized that GaussDB, founded on Huawei's 20-plus years of experience in databases, is aimed at meeting customers' high requirements for database intelligence and automation.

Dr. Ntarmos added that this effort entails boosting "self-driving" capabilities across key phases of the data pipeline that encompass database consulting, development, and operations and maintenance (O&M). During the consulting phase, GaussDB can automatically generate high-level designs (HLDs) by connecting large and small models to improve solution design efficiency, for example. This benefit, he estimated, can shorten the solution design period from about two weeks to about two days.

In the development phase, GaussDB directly generates SQL statements through natural language-to-SQL and SQL-to-SQL capabilities, with automatic identification and improvement suggestions for poor SQL statements. In the O&M phase, GaussDB automates inspection, report generation and fault demarcation while providing handling suggestions for defective SQL statements.

Dr. Ntarmos explained that the database's compatibility with the common syntax used in popular commercial databases makes it a migration solution that minimizes manual intervention and migration expenses. GaussDB has been adopted in the core systems of major banks, joint-stock banks, and insurance securities companies in China, as well as major Huawei Cloud customers in Brazil and Thailand.

Huawei Cloud focuses on AI as key strategy

Jacqueline Shi, President of Huawei Cloud Global Marketing and Sales Service, lauded AI's compelling potential to reshape almost everything. "At Huawei Cloud, AI is a key strategy," Ms Shi stated. "You can get comprehensive AI solutions with us, such as AI computing power solution and Pangu models. Huawei Cloud is one of the fastest growing cloud providers. We hope to bring leading-edge technologies, the best local service, and more and better choices to our customers."

"We are witnessing a new level of intelligence, driven by foundation models and generative AI," echoed Mr. Zhang. "It's predicted that by 2026, more than 80 percent of businesses will harness Artificial Intelligence Generated Content (AIGC) in production, transforming 70 percent of design and development work. By 2028, 75 percent of software engineers will have AI assistants by their side, up from just 10 percent at the start of 2023."

"Customers want collaborative heterogeneous computing architecture, cloud-native compute with superlative performance, mass data storage, security compliance, lean governance and flexible deployment," added William Fang, Chief Product Officer at Huawei Cloud. "It is only by integrating AI with cloud that progress in intelligence is possible."

To help customers fastrack their AI journey, Mr. Fang highlighted large language models, big data and computing power as key success factors in the development of the Huawei Cloud infrastructure.

More than these, even as production, service paradigms and business models for traditional applications are being redefined, accelerating intelligence while aligning AI's vast potential and implementation with business objectives requires systematic innovation.

To this end, Mr. Zhang proposed a two-pronged strategy. "AI for Cloud uses AI and foundation models to elevate your experience," he said. "They are reshaping industry applications and our own cloud services, including software development, digital content production, and more. Cloud for AI makes AI adoption seamless and efficient. Architectural innovation, AI-native storage, and data-AI convergence empower you to train and use AI like never before."

The systematic innovation and industry practices of Huawei Cloud encompass technical development in system architecture, compute, storage, database, big data, among others, as well as Pangu models for automotive, weather, virtual human, R&D, and more.

Systematic AI-business alignment

Explaining the Cloud for AI thrust, Mr Zhang said, "Foundation models and their applications are the most complex software and hardware systems to date. For example, we've discovered new challenges in compute. That's why we will shift from a CPU-centric architecture to a heterogeneous peer-to-peer cloud architecture, where varying compute resources work as equals."

Zhang added that cloud services is the key to achieving the hyperscale, stable and robust AI compute required by foundation models. Such compute far surpasses Moore's Law. Huawei Cloud has been helping telcos to build intelligent cloud infrastructure with its distributed QingTian architecture, AI-native storage, end-to-end security, and data-AI convergence.

The QingTian architecture removes the bottlenecks limiting the progress of large-scale AI compute clusters by enabling peer-to-peer full-mesh computing. A high-speed interconnect bus seamlessly integrates communications, resource management, and function calling into an optimal compute infrastructure.

"We have deployed three major AI compute centers and over 30 sub-centers in the Chinese mainland," said Mr. Zhang. "They run hyperscale clusters to support trillion-parameter models. Soon, our AI cloud service will go live in the Hong Kong Region to serve global customers."

Additionally, AI-native storage supports data-intensive training models. A three-pronged approach includes Expanded Memory Specification (EMS) memory service that stores petabytes of parameters with 220 TB ultra-large bandwidth and ultra-low latency down to the microsecond. This is complemented by Scalable File Service Turbo – a cache service for high throughput and concurrency of tens of millions input/output operations per second – and an Object Storage Service knowledge lake that stores training and inference data cost-effectively.

Huawei Cloud's end-to-end security protects industry customers' model runtime environments, training data, the models themselves, generated content, and applications.

Foundation models thrive on data, database

The proliferation of foundation models has also necessitated the need to provide quality data in a more efficient way to support model training and inference.

"We use LakeFormation, a unified data resource and metadata technology, to build a logical data lake from multiple data lakes or warehouses," said Mr Zhang. "This means that one copy of data can be shared by multiple data analytics engines and AI engines, without the need for data migration. Plus, our AI4Data engine makes the entire data governance process – from data integration, development, to quality and asset management – more intelligent."

At the Summit, Huawei Cloud also described how its AI-oriented innovations andextensive industry expertise of Pangu models – including the virtual human model – contributed to an AI-ready infrastructure that is reshaping many industries.

The Pangu models combine industry knowledge and large language model (LLM) capabilities to produce an expert assistant that could potentially make work more efficient and easier for industries, enterprises or individual users.

Unlike LLMs such as ChatGPT, Huawei Cloud's Pangu Models 3.0 features a three-layer decoupled architecture, said the company. The L0 layer consists of five foundation models, which provide general skills to power industry-specific applications. L1 contains industry-specific models that are trained using industry data, while L2 provides models for specific industry scenarios and tasks.

Mr. Zhang cited cases of where the Pangu telecom model helped telcos to automatically troubleshoot up to 90 percent of network faults in minutes while the Pangu R&D model enables developers to generate code with just one prompt and to test cases with just one click. In weather forecasting, Pangu has also been used to predict the path of a typhoon over a period of 10 days more accurately and within 10 seconds on a single server, he added.

Strikingly, the Pangu virtual human model boasts 95 percent lip sync accuracy for customer service and livestreaming. Virtual humans are widely used in fields such as e-commerce, news broadcasting, education and training.

In the automotive industry, Pangu automatically generates corner cases for all kinds of complex driving scenes, reducing the time it takes for autonomous driving models to learn new scenes. More models are in the pipeline. Pangu models and related cloud services can be deployed with trusted security and compliance on public cloud, dedicated zone or hybrid cloud.

With these announcements at the Summit, Ms Shi pressed home Huawei Cloud's plans to offer an open ecosystem of cloud services and new businesses. The company collaborates with customers and partners to create scenario-based solutions for vertical industries while it continues to expand KooVerse, Huawei Cloud's infrastructure, globally for easy access and high performance.

This article was contributed by Huawei.

More about

More about

More about

TIP US OFF

Send us news