A new class of cloud and datacenter infrastructure is emerging into the marketplace.  This new infrastructure element, called a DPU/IPU, or Data Processing Unit, takes the form of a server hosted PCIe add-in card containing one or more ASIC’s, usually anchored around a single powerful SoC device.  The DPU/IPU has its roots in the evolution of Smart NIC devices but separates itself from that legacy in important ways.  While Smart NICs are clearly part of their host node’s compute system and exist to closely interact with and offload node hosted applications, the DPU/IPU dispenses with this secondary role.  Instead, the DPU/IPU presents itself as a complete compute system unto itself with an independent software stack, network identity, and provisioning capabilities.  The DPU/IPU is fully capable of hosting its own applications using either embedded or orchestrated deployment models.

The unique capabilities of the DPU/IPU are disruptive because they allow for key infrastructure functions and their associated software stacks to be completely removed from the host node’s CPU cores and to be relocated onto the DPU/IPU. The host cores become the exclusive domain of the application workloads and are ideally managed by a DevOps team.  The infrastructure workloads, now executing on the DPU/IPU, are independently managed by the NetOps and SecOps teams.

Feature Overview

• Running network function on the DPU

Goals

CPUs are not getting ‘faster’ anymore in the face of the increasing need for speed, Network, Data and AI requirements, as well as concerns over power-consumption. Micro services and service meshes are providing agility but the cost of pack movement is taking more CPU resources.

The response to these challenges is the new system architecture exemplified by the Bluefield 2 DPU. The DPU in addition to the domain-specific hardware function and acceleration have their own compute capabilities - their own independent CPU.

Use cases

Acceleration and offloading of whole software subsystems rather than just specific tasks.

• This offers clean compartmentalization of the hardware accelerated platform, addressing the mismatching abstraction layers that compound the already exploding complexity of software stacks. 
• Essentially this means moving to higher-level APIs and abstractions at the service level.
• It matches the hardware architecture to the concepts of orchestrating encapsulated services that Containers and Kubernetes generalized on the software side.
• You could look at this as the ‘containerization’ of domain-specific hardware. 

Personas

• IT Management: Responsible for designing and providing an enterprise IT environment to best serve the core business. From Red Hat and NVIDIA perspectives, this IT environment could be either a Red Hat OpenShift or a Red Hat Enterprise Linux running on bare metal servers with NVIDIA BlueField DPUs, deployed across data center, edge, and public clouds.
  Key influencers
• IT Architects/Strategists - Responsible for identifying the most cost-effective, enterprise-grade IT infrastructure solutions to best serve the core business.
• IT Ops - Responsible for deploying, operating and maintaining the IT environment as well as delivering the highest levels of serviceability, availability, resiliency.
• Security Architects - Define and implement best security practices to safeguard various business-critical systems. 
• DevOps/DevSecOps Architects - Responsible for deploying and maintaining enterprise applications in a secure manner through automation and via collaboration with software developers and IT operations to ensure the highest levels of data privacy, integrity and reliability. In many businesses, DevSecOps and DevOps use similar tools and/or are coming closely together in an aim to enhance security,  streamline processes, and increase efficiency. 
  Primary target verticals: Financial Services, Telco, Cloud Service Providers

Other potential target verticals: Insurance, Government, Manufacturing, Healthcare, Life sciences, Higher Education, Energy