As organizations strive for digital transformation, they are in reality seeking to reinvent their business, modernize their processes and push the boundaries of existing IT infrastructure. To address that last point, we frequently see customers exploring alternatives to CPU-centric system architectures, where software running on a central CPU directly controls a set of HW peripherals that offer static functions and/or acceleration capabilities. Supporting this trend is a new class of devices, evolved from SmartNICs but lacking cohesive standards even when it comes to naming - NVIDIA and Marvell call their offerings Data Processing Units (DPU) while Intel refers to their technologies as Intelligent Processing Units (IPUs).
Red Hat firmly believes that this transformational change in compute architecture is only possible through open source and broad partner collaboration. No individual company or a "walled garden" approach could steer these efforts in the right direction, essentially requiring the creation of an open source-based architecture and the associated ecosystem behind it. Today, as a culmination of our efforts to structure a community around these new technologies, we are announcing that Red Hat is joining the Open Programmable Infrastructure (OPI) Project under The Linux Foundation, as a founding Premier member.
We believe there is an exciting opportunity to develop a fully programmable open infrastructure model across software stacks and DPU/IPU-like hardware devices. Red Hat has been driving the formation of this community for almost a year, closely collaborating with several industry partners and customers, such as Dell Technologies, F5, Intel, Keysight Technologies, Marvell, NVIDIA, and others. The OPI Project aims to explore and expand the concept of programmable infrastructure using community-driven, standards-based, open ecosystem for next generation architectures and frameworks based on DPU/IPU-like technologies.
The common trait for the emerging generation of DPU/IPU devices is that they employ an easily programmable multi-core CPU, a state-of-the-art network interface(s), and a powerful set of networking, storage and security accelerators that can be programmed to perform multiple software-defined, hardware-accelerated functions.
As the above illustration suggests, DPU/IPU-like devices provide a platform to enable a broad range of services across management, network and storage domains and represent the first wave of devices that support a programmable infrastructure concept where intelligent subsystems are matched by hardware components and orchestrated using Kubernetes-like patterns. As this segment matures, we are likely to see additional form-factors, more evolved architectural implementations and new use cases.
Customers and partners alike, value choice in their technology implementations, and choice is a key benefit of open source solutions. They need well-balanced and optimized infrastructure in their hybrid cloud implementations rivaling those that hyperscalers, such as Amazon with their proprietary Nitro architecture, offer. This new class of DPU/IPU-like devices in combination with open source software, like general purpose Linux operating systems and cloud-native Kubernetes, democratize access to disaggregated composable systems and programmable infrastructure.
Given Red Hat’s role in delivering open source-based enterprise-class software that brings greater consistency across multiple hardware implementations, we are working closely with partners and customers to help this new class of devices adhere to key ecosystem standards. Standardization is foundational to the success of this vibrant ecosystem and that necessitated the establishment OPI project that aims to provide clear guidelines for defining the applicable category of devices, create vendor agnostic framework and architecture for DPU/IPU-based software stacks, define or re-use a set of common APIs and provide implementation examples to validate those architectures and APIs with adequate performance levels.
However, we can not do that alone. While it is still in its infancy, the OPI project needs your help in establishing common frameworks and architectures that are applicable to any vendor’s hardware solution for this new ecosystem.
Whether you are a customer that wants to easily integrate Red Hat’s software and standardize their infrastructure across various hardware architectures and solutions or a partner looking to develop software solutions that will run on a multitude of hardware implementations, we encourage you to participate in the OPI Project.
About the authors
Kris Murphy is a Senior Principal Software Engineer in Red Hat’s Office of the CTO leading the Emerging Tech Computational Infrastructure team. Kris' team focuses on emerging hardware and architectures that may influence Red Hat's future products. Current focus areas include ARM based edge devices, DPUs/IPUs, and composable compute. The team works closely with Red Hat product teams, open source communities, partners and customers.
Yan Fisher is a Global evangelist at Red Hat where he extends his expertise in enterprise computing to emerging areas that Red Hat is exploring.
Fisher has a deep background in systems design and architecture. He has spent the past 20 years of his career working in the computer and telecommunication industries where he tackled as diverse areas as sales and operations to systems performance and benchmarking.