AI File Storage (AFS) is a scalable parallel file storage system launched by SenseCore, which is optimized for GPU-based high-performance computing scenarios and characterized by providing sub-millisecond-level data access, meeting high-IOPS and high-throughput data read and write requests, and applying to high-performance computing scenarios such as AI training, autonomous driving, and video rendering.
Realize in-depth optimization for AI application scenarios by using a distributed parallel architecture and supporting a throughput of tens of GB, sub-millisecond-level latency, and millions of IOPS
Reach up to 99.9% availability of architecture design and stable performance for tens of billions of small files without attenuation
Support online elastic scalability to meet the linear growth of capacity requirements and performance
Realize in-depth optimization for AI application scenarios by using a distributed parallel architecture and supporting a throughput of tens of GB, sub-millisecond-level latency, and millions of IOPS
Reach up to 99.9% availability of architecture design and stable performance for tens of billions of small files without attenuation
Support online elastic scalability to meet the linear growth of capacity requirements and performance
Support the POSIX file protocol, and efficiently execute high-performance AI computing programs without the need of adaptation and optimization;
Integrate with bare metal, container, and cloud lab computing platforms to facilitate support for the AFS file system; and support SenseCore container services and provide a high I/O capability for container services through the CSI interface;
Give full play to the advantages of concurrent access in a parallel I/O architecture, and maintain ultra-low latency of sub-milliseconds when the cluster achieves high throughput through horizontal scaling.
AFS is deeply integrated with AI application scenarios, supports access to massive small files, and improves AI training performance.
It is connected with various computing clusters to provide ultra-high throughput and ultra-high IOPS capabilities.
The high performance and elastic scalability of AFS strongly support the large capacity and high performance requirements brought about by customers’ business growth.
AFS is used in conjunction with AOSS. The data to be trained is imported from AOSS to AFS, and model training is performed in AFS. After the training is completed, the data is archived to AOSS for long-term storage.
AFS also provides high throughput and high IOPS to improve computing efficiency and reduce the overall business cost.
The architecture supports elastic scalability to achieve linear growth in capacity and performance.
AFS is deeply integrated with AI application scenarios, supports access to massive small files, and improves AI training performance.
It is connected with various computing clusters to provide ultra-high throughput and ultra-high IOPS capabilities.
The high performance and elastic scalability of AFS strongly support the large capacity and high performance requirements brought about by customers’ business growth.
AFS is used in conjunction with AOSS. The data to be trained is imported from AOSS to AFS, and model training is performed in AFS. After the training is completed, the data is archived to AOSS for long-term storage.
AFS also provides high throughput and high IOPS to improve computing efficiency and reduce the overall business cost.
The architecture supports elastic scalability to achieve linear growth in capacity and performance.
Help you achieve new breakthroughs in business with professional AI solutions and advanced AI products
