To improve the full ecological service of AI computing power and solve the industry pain points of data storage and high-speed access of large-scale AI clusters, Pro Tech Data independently developed a parallel file system suitable for large-scale AI clusters, building a core technological barrier of “top-tier hardware + self-developed software”, improving the overall efficiency of the computing power cluster, and providing stable and efficient data support for AI businesses.
- High-Performance Reading & Writing: Adopting a distributed parallel architecture, it supports 100GBps-level throughput bandwidth, sub-millisecond access latency, and 10-million-level IOPS concurrent performance, perfectly adapting to the high-throughput and low-latency needs of large-scale AI training and massive data processing.
- High Reliability: The architecture design has an availability of 99.9%. In the scenario of 10 billion small files, the operation and read-write performance remains stable without attenuation, ensuring data security and business continuity.
- Elastic Scalability: Supports online elastic scaling, which can quickly achieve linear growth of capacity and throughput performance, perfectly adapting to the business needs of expanding AI cluster scale.
- High Compatibility: Optimize the distributed parallel architecture of metadata and data, support concurrent access by thousands of servers at the same time, and adapt to GPU multi-machine and multi-card training scenarios; support POSIX file protocol, so AI high-performance computing programs can be efficiently executed without secondary adaptation.
Solve the core pain points of data storage and high-speed access of large-scale AI clusters, provide stable, efficient and safe data storage and scheduling support for customers’ AI businesses, maximize the value of computing power hardware, improve the R&D and operation efficiency of AI businesses, and help customers quickly promote technology landing.
