With the rise of the cloud era and the proliferation of the cloud throughout PCs, workstations, and computing services, the cloudification of high-performance computing is an inevitable trend. Cloud computing enables all scientific data to be processed in clouds, meeting enterprises' computing requirements, reducing the required heavy assets and maintenance workloads.
Scientific computing has evolved from a single-node system to a cluster-based system. Many scientific enterprises have looked to build their own HPC clusters. However, they face many problems with the construction and use of HPC clusters, including:
High construction costs and long rollout periods: Building HPC clusters requires a heavy investment of both time and money. This wastes resources and often fails to meet enterprises' requirements for service flexibility.
Heavy maintenance workloads: Maintaining HPC cluster consumes enterprise manpower and materials. As a result, enterprises cannot focus on their core services.
High OPEX: Operation costs for self-hosted HPC clusters are high, such as rent, electricity, water, and system management.
New technology emerges and evolves so quickly that enterprises can barely keep up with computing requirements.
Asset-heavy Operations and Sluggish Construction
Resources are exclusively planned and purchased based on the demand during peak hours, resulting in different enterprises and departments purchasing the same resources. It often takes several months to complete service construction, which cannot keep up with business growth.
Heavy Maintenance Workloads
As service develops and node size increases, maintaining a timely service response with zero downtime requires huge maintenance workloads, which distracts enterprises from their own service development.
Rapid Hardware Iterations
New technologies and frequent hardware updates make enterprises with numerous outdated devices unable to keep up with computing requirements.
Operational costs for managing self-hosted HPC clusters are high. Those costs include rent, electricity, water, and system management.
Scientific computing refers to computations conducted during both scientific research and engineering research endeavors, such as particle collision analyses or Monte Carlo calculations. Huawei's scientific computing solution provides excellent, convenient, and secure computing services.
Scalable and High-performance ECS
Provides various types of ECSs and BMSs to meet different HPC application requirements for computing and memory capacity. ECS types include H1 (high computing I, which uses a 10GE network), H2 (ultra-high computing II, which uses a 100 Gbit/s InfiniBand network), P1 (Artificial Intelligence), and M1 (memory optimization type).
Provides excellent computing performance using dedicated physical servers, avoiding loss due to virtualization and offering industry-leading performance. It allows you to automatically provision BMSs on the management console, fully meeting scalability and flexibility requirements for HPC services.
High-performance Lustre Parallel File System
Uses optimized and dedicated ECSs and EVSs to provides up to 4.5 Gbit/s throughput for each OSS and up to 2 Tbit/s throughput for all. Each file supports up to 32 PB of capacity. Lustre file system can fully meet the file system requirements of HPC.
Distributed and Scalable Block Storage Service
Allows EVSs to be attached to each ECS on demand and offers storage space for HPC computing and storage nodes. Each disk provides up to 32 TB of capacity, 20,000 IOPS, and 350 MB/s throughput.
Provides high-performance, reliable, and simple computing services to meet the computing capabilities required by scientific research.
- Excellent performance: 100 Gbit/s InfiniBand computing network, 3.2 TB SSD, the largest local cache disk for cloud services, and a fat node with 96 cores and 2 TB memory
- Rich instances: support for the deployment of ECSs and BMSs, meeting deployment requirements for different applications
- Open architecture: OpenStack-based architecture
- Comprehensive protection: security protection for 7 layers, including access, transmission, infrastructure, network, virtualization, data, and management layers, layer-3 Anti-DDoS, VPC, and IAM
Elastic Cloud ServerElastic Cloud Server (ECS) provides scalable, on-demand cloud servers for secure, flexible, and efficient application environments, ensuring reliable, uninterrupted services.
Bare Metal ServerBare Metal Server (BMS) provides dedicated physical servers in single-tenant environments. It provides excellent computing performance and data security for core databases, key application systems, and high performance computing. It also offers the high scalability of a cloud-based service.
Image Management ServiceImage Management Service (IMS) allows you to easily manage images. You can use a public image or create a private image to rapidly create or replicate Elastic Cloud Servers (ECSs) in batches.