Layerscape edge DPU for cloud data
SolidRun in Israel has developed a software defined data processing unit (DPU) card for edge and cloud data centres.
The half height, half length PCI Express 3.0 SolidNet card is based on NXP’s Layerscape 16 core ARM processor. This makes the card suitable for edge processing applications as well as cloud data centres.
The LX2162A-based board consumes up to 25W of power and provides the flexibility to implement software-defined networking (SDN), networking security and acceleration functions based on virtIO, DPDK and NVMe standards for next-generation cloud-scale computing. This is possible on bare metal designs for more efficiency, rather than using additional layers of software.
“With the increasing demand for performance in data centers, securing the infrastructure and scaling performance becomes a major challenge,” said Mordi Blaunstein, VP marketing and sales at SolidRun. “SolidRun’s DPU makes it possible to isolate the servers from the infrastructure on the node level, abstracting the infrastructure based on open standards, thus making it higher performance for about the same cost as a standard smart NIC.”
The card allows 100 percent bare-metal and virtualized/containerized Infrastructure as a Service (IaaS) to fully off-load the host server CPU’s hypervisor, handing network and NVMe functions to the DPU, creating physical isolation between the server compute resources and infrastructure.
This provides service offload capabilities for network management, storage, security abstractions to hide advanced topologies from applications, SDN overlays, mesh, NVMeoF, network HSM, virtual private cloud (VPC) and more.
The dedicated DPU boosts application performance and means these data and network functions no longer compete with application workloads for server cores, memory or storage resources, and benefit from improved server security with independent out-of-band management capabilities, secure boot, and root of trust support.
The software defined operation also supports workloads at the cloud edge to boost efficiency by maximizing the utilization of the host CPU for running the required application/service algorithms, while the DPU executes the basic function per application instructions. For example, DDoS attack filtering can be executed by the DPU per filtering tables built and maintained by the application running on the host server CPU.
“DPUs provide an effective way to drive more efficient resource utilization and reduce operating costs throughout the data centre,” said Imran Yusuf, director of hardware ecosystem, Infrastructure Line of Business at ARM. “Offloading applications that historically ran on server CPUs to high performance and efficient ARM-based CPUs is benefiting a wide range of applications from AI/ML, databases, storage and security tasks, and microservices. SolidRun’s Software-Defined DPUs will provide a streamlined user experience taking advantage of the robust ARM software ecosystem.”
Other ARM-based SoC options will be announced later says SolidRun.
SolidNET products are compatible with Linux-based SDN software applications, including DPDK, which provides data plane libraries and network interface controller polling-mode drivers for offloading many networking services and functions. For example, TCP segmentation from the operating system kernel to processes running in the DPU user space.
Other articles on eeNews Europe
- AI ‘unicorn’ Graphcore set to cut jobs
- Quantinuum doubles quantum volume of its computer
- Nearly 900 organisations use LTE/5G private networks
- MicroLED AR smart glasses uses MEMS speakers
- £4m projects to clear space debris