Driven by hybrid cloud services, AI, virtualization and scale-out storage requirements, IT professionals are challenged to deliver reliable, speedy networking performance at a price that delivers the highest return on investment. For any given workload, the right mix of performance, reliability, security, efficiency and cost is vital.
That’s why HPE is now offering the latest generation of NVIDIA Mellanox ConnectX-6 Dx Ethernet adapters to power its advanced cloud, enterprise and AI systems.
The adapters provide advanced, built-in capabilities that bring smart offloads and robust security down to the end-points of the network, delivering unprecedented performance and scalability while satisfying the most demanding workloads for today’s modern data centers.
Compatible with HPE ProLiant Gen 10 Plus servers (NVIDIA product number MCX623106AS-CDAT), these adapters provide dual ports of 100Gb/s connectivity, PCIe Gen4 support and up to 200Gb/s of aggregate throughput with QSFP56 optical transceivers in a low-profile PCIe form factor.
For the newest servers, 25Gb/s and 50Gb/s networking don’t provide enough throughput. And for virtualized, cloud and AI workloads, using a standard NIC doesn’t provide enough performance, efficiency or security.
ConnectX adapters have become the industry’s choice for secure and advanced cloud smart network interface cards (SmartNICs), accelerating mission-critical data-center applications such as security, virtualization, SDN/NFV, big data, machine learning and storage. HPE ProLiant Gen 10 Plus Servers with the new NVIDIA ConnectX-6 Dx adapter can enhance many solutions, including the following:
Resilient RoCE for AI, Big Data and Storage
Network-intensive applications like networked storage or cluster computing need a network infrastructure with a high bandwidth and low latency. RDMA over Converged Ethernet (RoCE) uses advanced Ethernet networking and offloads to reduce latency for applications requiring higher throughput and faster response times.
A key technology advancement in GPU-GPU communications has been GPUDirect RDMA, which feeds data directly to GPU memory without needing to make buffer copies or interrupt the host. ConnectX-6 Dx makes it easier to deploy GPUDirect over RoCE by enabling a direct RDMA path for data exchange between the GPU and other servers or storage.
Building on the Mellanox ConnectX tradition of industry-leading RoCE innovations, ConnectX-6 Dx also enhances the Zero Touch RoCE (ZTR) feature to enable more scalable, resilient and easy-to-deploy systems. ZTR allows RoCE-accelerated workloads to run on existing networks without requiring special configuration on the network (PFC and ECN are now optional), greatly simplifying deployments.
If a packet is dropped, a new ZTR option resends only the lost packet, reducing delays and improving throughput. These features make it easier and faster for HPE customers to accelerate Spark and storage using NVMe over Fabrics, as well as AI, machine learning and graphics workloads that use GPUs.
Advanced SDN and Virtualization
Virtualization is critical for enabling cloud data centers to deliver agility, flexibility and scalability. These benefits often come at the expense of performance and efficiency, as virtualization traditionally is performed by software that consumes CPU resources.
NVIDIA Mellanox SmartNICs feature ASAP2 (Accelerated Switch and Packet Processing) technology for vSwitch/vRouter hardware offload to deliver orders of magnitude higher performance vs. software-based solutions. ConnectX-6 Dx offers enhanced ASAP2 capabilities, supporting both SR-IOV and VirtIO in-hardware offload capabilities with millions of packet processing rules.
Additionally, NVIDIA ConnectX offers industry-leading DPDK performance on HPE servers, which yields higher performance, better security, easier management, and better transparency and troubleshooting. ConnectX-6 Dx includes advanced on-chip networking for traffic between VMs and containers, supporting efficient private and hybrid cloud deployments.
These innovations enable HPE servers with ConnectX-6 Dx adapters to deliver fast, efficient networking for virtualization and containers. It boosts SDN, NFV and next-generation firewall performance using software-defined, hardware-accelerated networking.
Best-in-Class Storage Offloads
With AI, big data and real-time analysis workloads, faster connections are necessary not just between servers but also to storage. Beyond just a faster connection speed, the new ConnectX-6 Dx adds a layer of innovation to enable more scalable, resilient and easy-to-deploy storage solutions. This includes acceleration and offloads for NVMe drives, NVMe over Fabric (NVMe-oF) and RDMA technology, plus optional Data Integrity Factor (T10-DIF) offloads and storage encryption for both data at rest and data in motion.
Recent enhancements to Microsoft’s Azure stack and Storage Spaces Direct (S2D) and VMware vSAN include implementation of high-speed RoCE transport for transferring large amounts of data with little CPU intervention. With ConnectX-6 Dx storage and networking offloads, the CPU is free to run more applications, which reduces the number of servers required to support a given workload, resulting in both capex and opex savings.
Advanced Security Features
In virtualized and cloud environments, perimeter-only security techniques are no longer adequate as they leave virtualized workloads, endpoints and edge devices vulnerable. In an era where privacy of information is key and zero trust is the rule, the ConnectX-6 Dx adapter offers advanced security features such as hardware root of trust (RoT) and connection tracking, enabling zero-trust security throughout the data center. ConnectX-6 Dx also enables a hardware-accelerated L4 firewall capability by offloading stateful connection tracking through ASAP2.
Secure boot supports a hardware RoT to protect against supply chain and other physical attack vectors, enabling a chain of trust for authenticating updates to adapter firmware. This blocks installation of rogue or corrupted firmware and ensures that the executing firmware is trusted.
5G for Time-Sensitive Networks – Time-Triggered Transmission Technology for Telco 5T for 5G
To support a transition to 5G networks, telco companies must address a longstanding problem of clock synchronization. This has to occur at the network edge, where the front-haul access network between radio resource units and baseband units requires an exact time synchronization for accurate management of radio resources and radio signal processing.
A breakthrough technology offered by NVIDIA in ConnectX-6 Dx adapters is the Time-Triggered Transmission Technology for Telco, known as 5T for 5G. It delivers incredibly accurate time synchronization across front-haul and mid-haul networks, providing telecommunications providers with higher performance, precise timing and reduced costs in their 5G rollouts.
ConnectX-6 Dx is the industry’s first network adapter to offer this precise time synchronization for eCPRI and provides a 5-in-1 solution: high-throughput, low-latency network connectivity, low power, a broad set of in-hardware acceleration capabilities, and 5T for 5G. It renders expensive and power-hungry FPGA devices unnecessary in front-haul networks.
For modern data center needs, HPE offers ConnectX-6 Dx network adapters for efficient storage and security that’s supported and integrated across its server and storage portfolio and to deliver faster, more efficient and more secure access to data for enterprise applications.