As organizations continue to push the boundaries of modern applications, AI workloads, and high-performance computing, networking performance becomes a critical factor in overall infrastructure efficiency. VMware Cloud Foundation (VCF) 9.1 introduces several significant enhancements to the NSX data plane, focusing on accelerating throughput, reducing latency, and improving scalability across both traditional enterprise and next-generation AI workloads.
In this third installment of our VCF 9.1 networking series, we'll explore the latest innovations in Enhanced Data Path (EDP), accelerated networking with NVIDIA SmartNICs, and performance improvements designed to help organizations extract maximum value from their infrastructure investments.
Driving Network Scalability with FPO Hardware Steering
One of the key enhancements in VCF 9.1 is the introduction of FPO Hardware Steering for NVIDIA accelerated networking adapters.
Modern data centers increasingly rely on hardware-assisted networking to handle growing volumes of east-west traffic. As virtual environments scale, the ability to efficiently program network flows becomes a critical performance consideration.
VCF 9.1 introduces an asynchronous queue-based hardware steering engine that significantly improves:
- Flow programming rates
- Infrastructure scalability
- Network processing efficiency
- Overall data path performance
This capability is available on a range of NVIDIA accelerated networking platforms, including:
- NVIDIA ConnectX-6 DX
- NVIDIA ConnectX-7
- NVIDIA BlueField-2
- NVIDIA BlueField-3
The enhancement benefits both:
- Uniform Passthrough (UPTv2)
- Enhanced DirectPath I/O (EDPIO)
By streamlining how flow entries are programmed into hardware, organizations can support larger-scale deployments while maintaining consistent networking performance across demanding workloads.
Uniform Passthrough (UPT): Balancing Performance and Flexibility
One of the most exciting additions in VCF 9.1 is the continued advancement of Uniform Passthrough (UPT) for accelerated networking environments.
Traditionally, organizations seeking maximum network performance often had to sacrifice operational flexibility, particularly when leveraging direct hardware access technologies. UPT changes that equation.
UPT delivers near line-rate networking performance using the familiar and widely adopted VMXNET3 interface, while offloading the data path directly to supported NVIDIA hardware such as:
- NVIDIA ConnectX-7
- NVIDIA BlueField-2 (NIC Mode)
- NVIDIA BlueField-3 (NIC Mode)
Why UPT Matters
A major advantage of UPT is that it eliminates the dependence on proprietary guest operating system drivers.
This provides several operational benefits:
- Simplified VM image management
- Easier operating system maintenance
- Consistent VM deployment models
- Reduced administrative complexity
Equally important, UPT maintains support for critical virtualization features, including:
- vMotion
- Workload mobility
- Guest operating system portability
Ideal Use Cases
UPT is particularly well suited for:
- Enterprise application workloads
- Database servers
- Business-critical virtual machines
- Large-scale virtualized environments
- High-performance applications that still require mobility
Key Benefit
Organizations gain high-speed networking performance while preserving the operational advantages that make virtualized infrastructure flexible and easy to manage.
In many environments, UPT offers an attractive "best of both worlds" approach by delivering accelerated networking without compromising day-to-day operational workflows.
Enhanced DirectPath I/O: Maximum Performance for AI and HPC
For organizations that require absolute networking performance, VCF 9.1 further strengthens support for Enhanced DirectPath I/O (EDPIO).
EDPIO provides virtual machines with direct access to underlying network hardware, allowing workloads to bypass traditional virtualization layers and achieve extremely low latency and high throughput.
Supported hardware includes:
- NVIDIA ConnectX-6 DX
- NVIDIA ConnectX-7
- NVIDIA BlueField-3 (NIC Mode)
Built for the Most Demanding Workloads
EDPIO is designed specifically for compute-intensive environments where every microsecond counts.
Typical use cases include:
- Artificial Intelligence training
- Machine Learning clusters
- High-Performance Computing (HPC)
- Large-scale distributed computing
- High-frequency trading platforms
GPUDirect RDMA Support
A particularly powerful capability enabled through EDPIO is GPUDirect RDMA.
This technology allows GPUs to communicate directly across the network with minimal CPU involvement, enabling:
- Faster data transfers
- Reduced latency
- Improved GPU utilization
- Accelerated distributed AI training
As AI workloads continue to expand across enterprises, technologies such as GPUDirect RDMA become increasingly important for reducing training times and maximizing infrastructure efficiency.
Key Benefit
EDPIO delivers the highest available performance by providing hardware-level access, making it the preferred networking architecture for organizations running mission-critical AI, HPC, and ultra-low-latency applications.
Enhanced Data Path (EDP) Continues to Evolve
VCF 9.1 introduces numerous improvements to Enhanced Data Path Standard, further improving resiliency and performance across NSX networking deployments.
Improved Deployment Experience
Maintaining uptime during infrastructure maintenance is always a priority.
VCF 9.1 introduces an updated orchestration sequence that helps maintain data path availability in environments utilizing multiple uplinks during planned maintenance activities.
Benefits include:
- Reduced service disruption
- Improved operational continuity
- More resilient maintenance workflows
Lower Latency and Higher Throughput
Several enhancements contribute directly to improved networking performance.
New optimizations include:
- Posted interrupt support for lower VMXNET3 latency
- Enhanced multicast Fast Path processing
- VMKAPI checksum offloading
Together, these improvements help reduce processing overhead while increasing overall throughput and responsiveness.
Organizations running latency-sensitive workloads will particularly benefit from these enhancements.
Expanded Feature Support
VCF 9.1 also broadens EDP feature capabilities.
New functionality includes:
- SR-IOV support for EDP Standard
- RDMA feature parity with pkthandle
- Large Receive Offload (LRO) support for Antrea deployments
These additions provide greater deployment flexibility and ensure organizations can leverage advanced networking technologies across a wider range of use cases.
Platform-Level Optimizations
Networking performance improvements extend beyond the data path itself.
VCF 9.1 introduces platform-specific optimizations for:
vSAN Environments
Enhanced support for:
- Multi-world VMKnics
- Improved storage network efficiency
- Better vSAN networking performance
AMD-Based Platforms
Additional optimizations include:
- Translation Lookaside Buffer (TLB) enhancements
- Improved memory efficiency
- Better packet-processing performance
These refinements help customers maximize performance from existing hardware investments while improving overall workload efficiency.
GRE Traffic Scaling for Cloud Service Providers
Cloud Service Providers and large-scale multi-tenant environments often rely heavily on GRE-based overlay networking.
VCF 9.1 introduces GRE vNIC RSS hashing enhancements for NSX Edge Node virtual appliances.
These improvements enable:
- Better traffic distribution
- Improved CPU utilization
- Increased traffic scale
- Enhanced network throughput
By optimizing Receive Side Scaling (RSS) for GRE traffic processing, Edge Node appliances can better handle large volumes of encapsulated traffic commonly found in service provider and large enterprise environments.
For organizations managing significant overlay networking deployments, this enhancement helps improve scalability without requiring additional infrastructure resources.
Final Thoughts
The networking enhancements introduced in VMware Cloud Foundation 9.1 demonstrate VMware's continued investment in delivering cloud-scale performance across both traditional enterprise workloads and emerging AI-driven environments.
From FPO Hardware Steering and Uniform Passthrough to Enhanced DirectPath I/O and the latest Enhanced Data Path optimizations, VCF 9.1 provides organizations with multiple options to match networking performance to workload requirements.
Whether you're running business-critical applications that require operational flexibility, large-scale cloud environments demanding greater scalability, or AI clusters that need every ounce of available performance, VCF 9.1 introduces the tools and technologies necessary to meet those demands.
As AI, machine learning, and data-intensive applications continue to reshape infrastructure requirements, these networking enhancements position VMware Cloud Foundation as a powerful platform for supporting the next generation of enterprise workloads.
Stay tuned for Part 4 of this series, where we'll explore additional networking innovations and capabilities introduced in VMware Cloud Foundation 9.1.
