Netris has released version 4.3.0 of its networking software, introducing advanced automation, abstraction, and multi-tenancy features specifically tailored for NVIDIA Spectrum-X, a platform designed for GPU-based AI cloud environments. This update builds upon Netris’ existing capabilities used by cloud operators in CPU cloud deployments, now bringing these functions to GPU cloud providers. The integration with NVIDIA Spectrum-X, which includes Spectrum-4 switches and BlueField-3 SuperNICs, enables optimized east-west traffic within data centers, significantly improving generative AI performance by 1.6 times over conventional Ethernet fabrics. This release marks a milestone in simplifying and automating AI-driven cloud networking infrastructure.
Netris’ solution for NVIDIA Spectrum-X automates switch-fabric management across day-0, day-1, and day-2 operations, facilitating easier deployment and ongoing management of high-performance AI networks. Unlike conventional approaches, which often rely on tightly integrated hardware and software with limited options, Netris disaggregates these components, allowing flexible configurations and support for a wide range of hardware. The software dynamically configures both east-west (AI-specific) and north-south (traditional data traffic) fabrics, optimizing network performance for GPU-intensive environments.
On day-0, Netris streamlines the setup process by automatically generating network blueprints through a Terraform module that takes into account NVIDIA’s best practices for Spectrum-X. This process simplifies network deployment, calculates optimized topologies, and configures switch fabrics based on GPU server counts. During day-1, users can leverage NVIDIA Air to create a digital twin of the network, simulating the design and ensuring performance before going live. The integration with NVIDIA Base Command Manager (BCM) further automates network bootstrapping and ongoing management.
Day-2 operations focus on continuous management of multiple fabrics and ensure that network isolation and multi-tenancy functions are applied dynamically, even as the network evolves. This includes support for monitoring features, alerting users to potential issues such as link errors or wiring mismatches. Advanced features such as Quality of Service (QoS), congestion control, and Adaptive Routing are also included to maximize AI-specific workload performance.
In addition to these core functionalities, Netris simplifies the complex task of multi-tenancy in AI clouds. Using a simple API, cloud operators can carve out isolated network clusters for different tenants, ensuring each tenant can only access their allocated resources. Netris abstracts the underlying complexity, automatically configuring the necessary fabric layers to meet these needs without requiring detailed manual inputs.
• Netris version 4.3.0 enables automation and multi-tenancy for NVIDIA Spectrum-X.
• Supports NVIDIA Spectrum-4 switches and BlueField-3 SuperNICs, optimizing AI network performance.
• Day-0 automation with Terraform module generates initial blueprints based on GPU server counts.
• Day-1 integration with NVIDIA Air allows for a digital twin simulation of the network.
• Day-2 operations include ongoing switch-fabric management and network isolation for multiple tenants.
• Dynamic configuration of east-west and north-south fabrics, supporting both AI and traditional traffic.
• Built-in monitoring alerts for network issues like wiring mismatches or link errors.
• API-driven network isolation and multi-tenancy with simple server lists and VPC configurations.
• Flexible, disaggregated approach compared to traditional integrated hardware setups.
• Includes advanced features like QoS, Adaptive Routing, and congestion control for AI-specific workloads.
• Full integration with NVIDIA Base Command Manager and support for InfiniBand environments.
• Designed for GPU-based AI clouds with east-west traffic optimization.
