Guizhou Meteorological Bureau: Computing at Full Power, Storing at Full Scale

Since 2023, with the implementation of policies such as the "Implementation Opinions of Guizhou Provincial People's Government on Promoting High-Quality Development of Meteorology in Guizhou" and the "Three-Year Action Plan for High-Quality Development of Meteorology in Guizhou (2023–2025)", Guizhou's meteorological industry has entered a fast track of development. However, limited computing power and near-saturated storage capacity could no longer weather the storm of exploding data volumes. Upgrading the computing and storage infrastructure became essential to clearing the bottlenecks. Against this backdrop, the Meteorological Bureau of Guizhou Province worked with Huawei, successfully delivering the three-phase HPC expansion project. The project establishes a high-performance, resilient, and scalable computing and storage foundation that powers more accurate forecasting and strengthens disaster prevention and mitigation.

Policy Support and Growing Services Fuel Demand for Computing Power

As the provincial-level meteorological authority under the China Meteorological Administration (CMA), the Meteorological Bureau of Guizhou Province is responsible for weather monitoring and early warnings, climate resource development, climate change research, and disaster prevention and mitigation. Its operations span ten internal departments, nine affiliated institutions, and nine municipal (state) meteorological bureaus, building a space-air-ground monitoring network. As forecasting services become more granular and update cycles continue to accelerate, data processing and storage are facing challenges:

The first challenge is data volume. Guizhou now produces localized forecasts at a spatial resolution of 1 kilometer—and as fine as 500 meters in some areas. Next-generation weather radars complete a volume scan every 6 to 10 minutes, while meteorological satellites capture regional imagery every 5 to 15 minutes. Nowcasting systems refresh two-hour forecasts every six minutes. Together with the continuous assimilation and analysis of multidimensional observational data, these workflows are driving explosive growth in the volume of meteorological data. A single historical data assimilation dataset reaches the petabyte scale, while storage deployed in the first two phases is already 75% utilized, making expansion imperative.

Performance presents an equally critical challenge. The real-time performance and accuracy of weather forecasts directly affect disaster prevention and mitigation. As forecast precision advances to the kilometer level, the computational demands rise exponentially and place demands on the system I/O performance. A storage architecture built on distributed I/O and parallel file systems is required to keep forecasting workloads running at full speed.

To meet these demands, the Meteorological Bureau of Guizhou Province and the Guizhou Provincial Big Data Bureau launched a three-phase project to modernize the province's meteorological computing platform, achieving 2,000 TFLOPS of peak computing performance and 10 PB of storage capacity. Completed in 2025, Phase III added 360 TFLOPS of compute capacity and 3 PB of storage, showcasing Huawei's expertise in HPC infrastructure.

Huawei HPC Storage Solution: Building a Tailored Data Foundation

To address the growing service challenges, Huawei delivered a tailored HPC storage solution built on OceanStor Pacific scale-out parallel file storage. Combined with joint optimization across compute, network, and scheduling software, the solution delivers breakthroughs in four critical areas: capacity, performance, compatibility, and reliability.

• Elastic scalability for the data deluge

Huawei OceanStor Pacific uses a fully symmetric scale-out architecture that enables capacity and performance to expand linearly. The storage system can be scaled seamlessly to thousands of nodes and exabyte-scale capacity, providing ample headroom for the meteorological demands of Guizhou. The automatic load balancing policy balances data and metadata across all storage nodes, eliminating metadata access bottlenecks. FlashLink technologies, such as intelligent stripe aggregation and I/O priority scheduling, provide high bandwidth and low latency, supporting the processing of massive volumes of gridded data.

• Ultra-high density design that keeps performance cool

To maximize limited data center space, Huawei has introduced OceanStor Pacific 9550 and 9950, engineered to deliver high capacity and performance. OceanStor Pacific 9550 high-capacity hardware features a high-cohesion structure that maximizes disk density. In addition, dual air channels and counter-rotating pressurized fans are used to improve heat dissipation. It is an industry-first, ultra-high-density storage device that fits seamlessly into 1.1 m cabinets. OceanStor Pacific 9950 high-performance hardware uses a full PCIe 4.0 architecture and Huawei-developed half-palm NVMe SSDs for all-flash performance. By deploying fewer devices, Huawei delivers higher capacity, more efficient data processing, and optimal space utilization.

• Service convergence that eliminates data silos

Huawei's solution supports the interworking of file, object, and HDFS unstructured data storage services, as well as multiple standard interfaces such as NFS, SMB, POSIX, MPI-IO, S3, and HDFS, enabling multiple services to share the same data. This solution delivers the high bandwidth and low latency required for HPC file services, enables flexible scaling for object storage, and meets the demands of big data analytics—all without data migration, significantly improving service efficiency.

• Proprietary parallel file system built for resilience

Powered by Huawei's proprietary parallel file system, the solution delivers high performance. A single all-flash node provides up to 90 GB/s of bandwidth and 65,000 OPS, while single-stream bandwidth reaches 8 GB/s. A single storage system efficiently supports various performance models, such as the large-file high-bandwidth model, random-I/O model, and small-file read/write model. The four-layer reliability design, including module-level hardware redundancy, node-level power failure protection and failover, system-level cross-node erasure coding (EC) and end-to-end data integrity field (DIF) verification, and solution-level comprehensive subhealth detection, ensures the resilience of meteorological data and protects against silent data corruption and data loss.

Turning Collaboration into Impact, Powering High-Quality Meteorological Development

With the completion of Phase III HPC expansion with Huawei, the Guizhou Meteorological Bureau has established a significantly more capable computing and storage platform. The additional 360 TFLOPS computing power and 3 PB of storage, combined with the infrastructure delivered in the previous phases, bring the platform to its target of 2,000 TFLOPS of peak computing performance and 10 PB of storage capacity. The upgraded infrastructure supports the local operation of national numerical weather prediction (NWP) models while removing storage bottlenecks.

At the service application layer, the high-bandwidth, low-latency storage has accelerated meteorological data assimilation and NWP computation, enabling faster updates and more accurate nowcasting. The enhanced platform strengthens early warning capabilities for severe convective weather—including heavy rainfall, hail, and lightning. The proprietary technologies and elastic scalability reduce the total cost of ownership (TCO) and provide the flexibility to support future AI model training and climate resource utilization. This aids the transformation from reactive disaster response to proactive service enablement.

The partnership rings together policy support and market collaboration, standing as a testament to Huawei's customer-centric philosophy. From expanding computing power to strengthening the data foundation, Huawei's HPC solution has accelerated Guizhou's journey toward a new era of meteorology. Looking ahead, the Meteorological Bureau of Guizhou Province and Huawei will deepen their collaboration, using Guizhou's status as a national hub of China's integrated computing power network to continuously advance meteorological computing capabilities. Together, they will safeguard the landscape and livelihoods of the province, building the climate resilience needed to secure Guizhou's sustainable economic future.