Coupled with Napatech software, the F2070X is the perfect solution for network, storage and security offload and acceleration. It enables virtualized cloud, cloud-native or bare-metal server virtualization with tenant isolation.

Powerful Intel^®-based Infrastructure Processing Unit (IPU)
The Napatech F2070X Infrastructure Processing Unit (IPU) is a 2x100Gbe PCIe card with an Intel^® Agilex^® AGFC023 FPGA and an Intel^® Xeon^® D SoC. The unique combination of FPGA and full-fledged Xeon CPU on a PCI card allows for unique offload capabilities. Coupled with Napatech software, the F2070X is the perfect solution for network, storage and security offload and acceleration. It enables virtualized cloud, cloud-native or bare-metal server virtualization with tenant isolation.

Customization on demand
The F2070X uniquely offers both programmable hardware and software, to tailor the IPU to the most demanding and specific needs in your network, and to modify and enhance its capabilities over the life of the deployment. It is based on the Intel Application Stack Acceleration Framework (ASAF) that supports the integration of software and IP from Intel, Napatech, 3rd parties and homegrown solutions. This one of a kind architecture enables hardware performance and the speed of software innovation.

Scalable platform
The Napatech F2070X comes in a standard configuration, and includes support for several combinations of Intel^® FPGAs, Xeon^® D processors, and memory. This enables tailored platform configurations matching requirements for specific use cases.

This IPU-based architecture has several major benefits

• First, the strong separation of infrastructure functions and tenant workload allows tenants to take full control of the CPU.
  • Guest can fully control the CPU with their SW, while CSP maintains control of the infrastructure and Root of Trust
• Second, the cloud operator can offload infrastructure tasks to the IPU. This helps maximize revenue.
  • Accelerators help process these task efficiently. Minimize latency and jitter and maximize revenue from CPU
• And third, IPU’s allow for a fully diskless Server Architecture in the cloud data center.
  • Simplifies data center architecture while adding flexibility for the CSP

Infrastructure Processing Unit

An IPU offers the ability to:

• Accelerate infrastructure functions, including storage virtualization, network virtualization and security with dedicated protocol accelerators.
• Free up CPU cores by shifting storage and network virtualization functions that were previously done in software on the CPU to the IPU.
• Improve data center utilization by allowing for flexible workload placement.
  Enable cloud service providers to customize infrastructure function deployments at the speed of software.

IPUs are architected from the ground up specifically to be an infrastructure control point. Intel^® IPUs provide the greatest level of security in a bare-metal hosted environment.

• Research from Google and Facebook has shown 22%^[1] to 80%^[2] of CPU cycles can be consumed by microservices communication overhead.
• Some of the key motivations to using an IPUs are:
  • Accelerate networking infrastructure by moving some of the workload onto the IPU this reduces the total server overhead.
• Reference Workloads are localized on the IPU. Performance is increased through:
  • Removing some of the PCIe latency between the host and the IPU.
  • Offloading SW functions to HW – For reference only: ZSTD Compression in HW can achieve L9 100Gbs @ 64KB packets and a compression ratio of 32.6%, a similar SW solution would consume 369 cores @ 1.8GHz achieve L7 @ 32.7% compression ratio.
  • Application-level HW optimization can be made in the FPGA and IPU.
• IPU are use reconfigurable and highly programmable allowing for customization of particular features and development and deployment on software timescales
• This IPU-based architecture has some major advantages:
  • First, the cloud operators can offload infrastructure tasks to the IPU. The IPU accelerators can process this workload very efficiently. This optimizes performance and the cloud operator can rent out 100% of the CPU to his guest which also helps to maximize revenue.
  • The IPU allows the separation of functions so the guest can fully control the CPU. The guest can bring its own hypervisor, but the cloud is still fully in control of the infrastructure and it can sand box functions such as networking security and storage.
  • The IPU can also help replace local disk storage directly connected to the server with virtual storage connected to the other network that will greatly simplify data center architecture while allowing tremendous amount of flexibility.

^[1] The reference “From Profiling a warehouse-scale computer, Svilen Kanev, Juan Pablo Darago, Kim M Hazelwood, Parthasarathy Ranganathan, Tipp J Moseley, Guyeon Wei, David Michael Brooks, ISCA’15” https://research.google/pubs/pub44271.pdf — figure 4.
^[2] https://research.fb.com/publications/accelerometer-understanding-acceleration-opportunities-for-data-center-overheads-at-hyperscale/