The flexibility offered by Software Defined Networks (SDNs) is appealing to network operators, since it allows the customization of network behavior for application-specific needs. In SDNs, flexibility comes at the cost of running the network on general-purpose hardware, reducing the performance that one can obtain from specialization. A recent trend in the industry is to use specialized hardware (ASICs and FPGAs) to combine the best of SDN programmability and flexibility with hardware execution efficiency. Programmable networking hardware allows both enhancing legacy protocols (e.g. adding monitoring to a traditional L2/L3 switching) and developing new protocols at a much faster pace.

To support programmability for network devices, P4 (www.p4.org) has been developed as a new programming language for describing how network packets should be processed on a variety of targets ranging from general-purpose CPUs to NPUs, FPGAs, and custom ASICs. P4 was designed with three goals in mind: (i) protocol independence: devices should not “bake in” specific protocols; (ii) field reconfigurability: programmers should be able to modify the behavior of devices after they have been deployed; and (iii) portability: programs should not be tied to specific hardware targets. P4 is the first widely-adopted domain-specific language for packet processing. Several vendors have developed FPGA-based implementations and, with the arrival of Tofino, there is already at least one domain-specific processor optimized as a compiler target for P4 programs – a processor with a small instruction set that can process one header per cycle. The P4 community has created – and continues to maintain and develop – the language specification, a set of open-source tools (compilers, debuggers, code analyzers, libraries, software P4 switches, etc.), and sample P4 programs, all with the goal of making it easy for P4 users to quickly and correctly author new data-plane behaviors. Specialized backend compilers and optimizers for vendor targets are built upon the open source framework. Using P4 and the open source tools, it is easy to prototype new ideas for networking hardware and applications in P4, by simply augmenting the compiler with support for the new hardware.

The goal of the tutorial is to introduce attendees to the domain of specialized hardware and programming tools for SDN. We discuss the basic operations in networking and how they have influenced the design of the P4 language, and of specialized, programmable networking hardware. We will show how the design goals of P4 language are met through examples of programs that can run on a variety of architectures. We provide several examples of applications (e.g. monitoring) that are enabled only by the combination of programmable hardware. We expect that, at the end of the tutorial, attendees will be familiar with the application domain and with a set of several implementations from different vendors that demonstrate various trade-offs. We aim to encourage researchers to consider the programmable networking devices area as one of the areas where domain-specific specialization is already moving into commercial devices and to contribute to the development of the P4 based ecosystem.

It is not hard to imagine a day in the near future where explaining the concept of driving would be analogous to explaining the concept of using a cassette player to play music today. The path to this scenario is paved by the rise of autonomous vehicles towards making our roads safer, and in doing so, redefining transportation as we know it.

NVIDIA’s DRIVE platform provides the foundation and the building blocks to enable development and a path to production for autonomous vehicle development. The hardware’s rich sensor input capability enables data acquisition for use cases such as neural network training and HD map generation and updates. The provided software accelerates the development of autonomous vehicle technologies such as perception, localization, and path planning. This platform provides a seamless transition for algorithms and tools created and tested on the development platform onto products that would demand the quality and safety certification supplied by our Tier 1 partners.

We will be covering the advancements in hardware which facilitate optimized computation for self-driving tasks, including the detection of objects/obstacles around the vehicle. Until recently, the perception of the world around the vehicle was primarily derived from hand-crafted computer vision algorithms. While these handcrafted algorithms were adequate for basic ADAS, it is simply impossible to manually write code for every possible scenario an autonomous vehicle (AV) might encounter. AV requires a new computing model which is deep learning. Deep learning algorithms help address the issues of robustness, accuracy, and scalability, and they have become more relevant through the advancements in the availability of big data, compute power, and accelerated frameworks.

The goal of this tutorial is to provide an overview of the autonomous vehicle landscape through NVIDIA’s platform and to highlight how deep neural networks are changing the autonomous vehicle landscape.