Part 1: Orientation & Setup

Welcome! We are excited that you want to learn InAccel.

Through this tutorial, we will guide you step-by-step on how to leverage InAccel toolset to ship and run your accelerator workloads in a simple, straightforward and efficient manner.

You will mainly learn how to:

1. Deploy a simple FPGA accelerated application
2. Develop your own FPGA accelerated application
   • Deploy your bitstreams
   • Accelerate your app
   • Deploy your app (Host/Docker/Kubernetes)

Download and Install InAccel-

This tutorial assumes that you have already installed InAccel. If you don't have InAccel FPGA Operator installed, choose your preferred deployment environment to download and setup InAccel.

Docker Kubernetes

Start the Tutorial-

Prior to the hands-on tutorial describing the exact steps on how to get started, we consider valuable to provide some motivation behind the necessity of an intelligent Bitstream Repository combined with Coral FPGA Resource Manager as well as some high-level details of their functionality.

Motivation-

The end of Moore's Law signified the shift from general purpose computing to tailor-made, specialized architectures and hardware. As a matter of fact, all major cloud providers have incorporated FPGAs into their cloud services.

While both CPUs and GPUs have developed a wide, well-established surrounding software ecosystem around their technologies, unfortunately FPGAs are unable to compete with the aforementioned hardware on the axis of software tools available. Currently, an FPGA application developer needs to be aware of low-level details of the underlying hardware. New layers of granularity need to be introduced to delegate the hardware details to accelerator designers while providing a typical software API to application developers.

At InAccel, we believe that wide adoption of hardware technologies requires significant levels of granularity and abstraction, and we are working towards that direction to mitigate the steep learning curve of developing, deploying and running accelerators.

Bitstream Repository Concepts-

Before we proceed with this tutorial, let's take a step back and elaborate on what a repository of bitstream artifacts represents and why do we need it in the first place.

What is a bitstream artifact?-

A bitstream artifact is a set of files capable of fully describing an FPGA design. InAccel introduces the notion of a specification file, where a bitstream developer can explicitly declare all the properties pertaining to the FPGA binary she is willing to deploy. Therefore, a compiled bitstream along with its specification file (metadata) comprise the bitstream artifact.

Why do we need a bitstream artifact?-

By accompanying each bitstream with a descriptive file that contains various metadata we are generating a high-level representation of the bitstream. Through this representation we achieve to decouple the background knowledge required by the accelerator designer and the application developers. In particular, the bitstream developer is responsible to create the specification file with all the required architecture and platform details. In contrast, the application developer needs only to be aware of the bitstream/kernel identifiers as well as the argument list of each accelerator to invoke them properly in her application.

Advantages of representing bitstreams as artifacts:

• Encapsulate every bit of information describing an FPGA design within a single file.
• Enable repository management, versioning capabilities and accelerator performance tracking (see below).
• Conceal low-level hardware details from the application developer allowing a universal, vendor agnostic (Intel / Xilinx) packaging format.
• Each artifact contains useful metadata for a wide range of utilities and tools.
• Representing self-contained entities as artifacts is a valuable software paradigm.

What is a bitstream repository?-

A bitstream repository is a central place in which an aggregation of the aforementioned artifacts is kept and maintained in an organized way. Protection (encryption/decryption) capabilities enable secure storage of your FPGA binaries to local and remote repositories with fine-grained access control policies. Moreover, the supplied metadata in each artifact simplify the process of maintaining snapshot/release artifacts, allowing continuous integration and delivery of your accelerated solutions through tailor made development pipelines. Now the evaluation and benchmarking of accelerators of different target vendors or versions is rendered trivial.

Finally, InAccel offers an end-to-end JFrog Bitstream repository solution to fully manage the deployment lifecycle of your FPGA binaries. More information can be found here.

What is bitstream deployment?-

Bitstream deployment is the process of storing your bitstream artifact into a bitstream repository. Once your bitstream is deployed, you can expect its kernels to be readily available from every application requesting accelerators from Coral. Additionally, through our CLI you can list its available accelerators along with many other useful information.

Coral FPGA Resource Manager Concepts-

Now that we addressed some major shortcomings of the specialized hardware industry and explained the main concepts of accelerator deployment that InAccel introduces, we continue by presenting InAccel's approach on FPGA application development deficiencies.

InAccel Coral is a scalable, reliable and fault-tolerant distributed acceleration system responsible for monitoring, virtualizing and orchestrating clusters of FPGAs. Coral also introduces high-level abstractions by exposing FPGAs as a single pool of accelerators to any application developer that she can easily invoke through simple API calls. Finally, Coral runs as a microservice and is able to run on top of other state-of-the-art resource managers like Hadoop YARN and Kubernetes.

To summarize, Coral has the following primary goals:

• Serve as a universal orchestrator for FPGA resources and acceleration requests.
• Improve scalability and maximize performance of deployed accelerators, ensuring the secure sharing of the available resources.
• Abstract away cumbersome parallel programming languages (like OpenCL) without compromising flexibility.
• Encompass bitstream management and protection capabilities.

Since we briefly covered the necessity of a universal FPGA resource manager as well as InAccel's approach on the issue, continue to Part 2 to run a sample application. You can find more information on Coral integrations and features here.