We use cookies to personalize your site experience.
Privacy policyMeet US at SC24
November 17-22 | Atlanta, GA | Booth 4603
Learn moreNetwork File System (NFS) is a distributed file system protocol that allows access to files over a network, ensuring seamless data sharing across diverse systems. It is widely adopted in both enterprise and high-performance computing environments for its simplicity and compatibility. However, traditional NFS implementations often face limitations in throughput and efficiency when dealing with modern AI, HPC, and large-scale data processing tasks.
By integrating software RAID solutions, particularly Xinnor's xiRAID, with NFS, we unlock a higher level of performance and scalability. Our RAID technology boosts NFS efficiency, reducing latency and improving data throughput, essential for AI workloads, virtualized environments, and more.
Optimized Performance
xiRAID dramatically enhances NFS performance by delivering higher throughput and IOPS, reducing bottlenecks typical in standard RAID setups.
Low Resource Consumption
With minimal CPU core usage (as few as 2-4 cores), xiRAID ensures you get maximum performance while keeping overhead low, a key advantage for both virtualized and bare-metal deployments.
Virtualized Environment Friendly
xiRAID is designed to excel in virtualized NFS environments, ensuring reliable and high-speed access to virtual block devices with the ability to saturate high-bandwidth network interfaces.
RDMA Support
xiRAID supports NFS over RDMA, allowing for faster data access with significantly lower latencies, essential for high-performance AI and machine learning tasks.
Scalability
Whether on-premises or cloud, xiRAID’s flexibility allows NFS environments to scale efficiently, ensuring optimal performance even as data demands grow.
When integrated with xiRAID, NFS delivers optimal performance for NVIDIA DGX systems, typically used in high-performance AI and machine learning. By leveraging NFSoRDMA, we ensure parallel access from multiple nodes and saturate a 400 Gbps network, achieving up to 50GBps of read throughput. This provides the necessary speed for high-demand tasks like model training and inference while maintaining data reliability and minimizing CPU load.
Numjobs | 1 | 4 | 8 | 16 | 32 |
---|---|---|---|---|---|
Sequential write, GB/s | 2 | 11.8 | 18.7 | 27.9 | 33.5 |
Sequential write, degraded mode, GB/s |
3.2 | 11.6 | 19 | 27.8 | 34.2 |
Sequential read, GB/s | 17.6 | 46.6 | 49.5 | 49.5 | 49.5 |
Sequential read, degraded mode, GB/s |
12.8 | 49.5 | 49.5 | 49.5 | 49.5 |
When simulating one drive failure with NFSoRDMA, no performance degradation is observed, meaning that DGX H100 client will not suffer any downtime.
Read the whitepaperIn AI and HPC, disaggregated storage solutions using NFS over RDMA have proven to meet the growing need for high throughput and low latency. Combining NFS with xiRAID Opus, our solution efficiently maximizes 400Gbps network interfaces while keeping resource consumption low. This setup ensures AI models can access large datasets at high speeds, making it ideal for virtualized environments with minimal CPU usage, meeting the performance needs of modern AI tasks.
For cloud deployments, xiRAID combined with NFS offers an excellent solution for handling high-throughput and scalable storage. We’ve optimized NFS for both synchronous and asynchronous operations, ensuring reliable performance for various cloud-based workloads. xiRAID ensures that NFS clients can access high-speed storage nodes efficiently, making it suitable for multi-tenant cloud environments where flexibility and scalability are critical.
Read the blog postAs data workloads continue to grow in complexity and scale, it’s crucial to have a storage infrastructure that is both adaptable and resilient. Xinnor’s integration of xiRAID with NFS ensures that your systems are prepared for future challenges, whether it’s handling the ever-increasing demands of AI training, large-scale HPC simulations, or multi-tenant cloud environments.
Our software RAID technology not only enhances performance but also simplifies deployment and management, making it easier to achieve high-speed, scalable storage across diverse applications. By choosing Xinnor, you’re investing in a solution that evolves with your needs, providing long-term reliability and performance.