Stormesh
Product Overview
Stormesh is a high-performance parallel file system developed to meet the demanding requirements of modern data-intensive workloads, delivering speed, scalability, and reliability for enterprise-grade environments.
Through innovations in metadata management and storage engine architecture, Stormesh achieves throughput of up to TB/s, tens of millions of IOPS, and consistently sub-millisecond latency. It is well suited for AI and machine learning, image processing, autonomous driving, EDA, scientific computing, and large-scale rendering workloads.
Built for enterprise-grade environments, Stormesh integrates advanced technologies such as SPDK (Storage Performance Development Kit) and RDMA (Remote Direct Memory Access) to maximize storage efficiency and accelerate data access.
Its flexible distributed architecture leverages node collaboration and intelligent data sharding to enable parallel storage and access across the cluster, significantly improving read and write performance. Combined with support for high-speed InfiniBand (IB) and RoCE networks, Stormesh delivers ultra-low-latency data transfer and a seamless user experience for performance-critical applications.
Stormesh supports multiple file access protocols, including POSIX, NFS, and SMB, providing reliable and high-performance file storage services for a wide range of enterprise workloads.
Product Advantages
High Performance
Stormesh features a proprietary POSIX-compatible client designed to fully utilize underlying hardware resources. Its intelligent data routing engine uses an efficient consistent hashing algorithm to shorten I/O paths and maximize storage efficiency.
- Up to TB/s-level throughput
- Up to tens of millions of IOPS
- Consistent sub-millisecond latency
- Supports demanding AI, HPC, and enterprise workloads at scale
High Availability
Stormesh uses distributed storage, multi-replica protection, and multi-node metadata redundancy to preserve continuity and durability across storage nodes, data, and metadata services.
- Storage nodes form a distributed storage cluster that ensures load balancing and continuous service availability
- Balanced data distribution and automatic failure recovery protect data durability and service continuity
- Every metadata node can process requests independently, eliminating metadata hotspots and improving metadata OPS
Elastic Scalability
Stormesh supports seamless online expansion without disrupting running applications. Storage capacity can be expanded dynamically as business requirements grow.
- Online expansion without disruption
- Up to 10 PiB per file system
- Scalable foundation for large AI and enterprise environments
Shared Access
Multiple hosts within the same project can mount and access a shared file system simultaneously, which fits collaborative and distributed workloads.
- Shared access across compute nodes
- No restrictions on the number of compute nodes that can access a file system
- Suitable for large-scale AI training, HPC, and distributed data processing
Ease of Use
With full POSIX compatibility, existing applications can integrate seamlessly without requiring code modifications. Standard file system management tools can be used directly.
- No code changes required
- Standard file system tools work directly
- No need to change established development or operational workflows
Use Cases
Stormesh is designed for data-intensive workloads that require high-performance shared storage, large-scale scalability, and concurrent access across multiple compute nodes.
Different deployment scenarios may require different performance profiles. The following are recommended use cases for Stormesh.
Large Language Model (LLM) Training
- High-concurrency data access
- Fast checkpoint storage and retrieval
- Low-latency dataset streaming
- Shared storage for GPU clusters
Autonomous Driving
- Large-scale sensor data processing
- High-speed data ingestion and transmission
- Simulation and model training workloads
- Distributed data management
Rendering & Visual Computing
- Simultaneous access to large media assets
- Support for hundreds or thousands of rendering nodes
- High-bandwidth file access
- Accelerated production pipelines
Electronic Design Automation (EDA)
- Efficient processing of massive volumes of small files
- High-performance simulation workloads
- Time-series analysis and verification tasks
- Scalable storage for chip design environments