High performance parallel file storage
Huacun Zhigu TGStor 9800 is a parallel file storage product designed for
high-performance data intensive processing workloads. Based on a scalable
system architecture, it consists of two modules: File Engine and Smart Pool.
Benefiting from core technologies such as high-performance parallel clients,
end-to-end full link RDMA low latency network optimization, discrete data
aggregation, and minimalist IO software stack for data throughput, a single
TGStor 9800 can achieve a maximum bandwidth of 70GB/s, 3 million IOPS,
and system performance can be linearly scaled up to TB level, achieving the
ultimate throughput capability required for full speed computing power operation
and maximizing data value.
TGStor 9800 is suitable for AIGC large model training, scientific computing in fields
such as meteorology, genetics, bioinformatics, astronomy, as well as high-performance
computing (HPC) and high-performance data analysis (HPDA) scenarios in industries
such as CAE simulation/GIS remote sensing/petroleum geophysical exploration.
KM and description
| Advantage Category | Specific Technical Features | Core Outcomes |
|---|
| Extreme Performance | Parallel architecture, reading data directly from multiple data nodes in parallel | Achieve low latency and high throughput |
| Intelligently identify small files/small IOs, and aggregate scattered data | Improve small file/small IO throughput performance |
| Innovative DPU card to accelerate protocol offloading and bypass data to disk | Increase read bandwidth by 30% |
| Disk scheduling algorithm to optimize CPU processing efficiency | Reduce data read/write latency, improve performance by 30% |
| achieving large IO metadata write caching, with data written directly to disk | Increase write bandwidth by 30% |
| Extreme Scalability | Both data and metadata can scale horizontally, single file system namespace | Capacity can scale linearly up to EB level |
| Combination of file index and intelligent disk frame pull | Independent expansion of each part, flexible adaptation to requirements |
| Linear expansion of system performance | Scalable to TB level, meeting storage needs for large-scale computing power |
| Intelligent data tiering and flow capabilities | Support parallel data exchange with third-party object storage/file storage and other massive data storage |
| Security & Reliability | Disk protection: NVMe SSD enterprise-grade dual-port hard disks, SSD global wear leveling and global reverse wear leveling, LDPC+FSP algorithms, in-drive DIF | Ensure disk data security and durability |
| Architecture protection: Key component redundancy, end-to-end high-availability architecture, millisecond-level fault switching, online replacement of controller/module faults, uninterrupted host routing | Uninterrupted business, zero IO failures |
| System protection: Erasure code data fragmentation algorithm, EC reconstruction | Rebuild time per TB shortened to 15 minutes, ensuring no drop in data redundancy |
| Health management: Hard disk health and network health management | Predict faults, avoid slow disks and network jitters from dragging down overall performance |
Technical Specifications
