Part 2: Sample application

Now that we have discussed about the main FPGA world concepts, let’s run a simple vector addition accelerated application. Depending on the deployed environment, we are going to either use Docker or Kubernetes instructions below:

DockerKubernetes

Step 1 Install docker-compose-plugin:
To run the sample application you will need to have installed the docker compose plugin. If docker compose is not installed in your system please follow the instructions below:

AptYum

sudo apt docker-compose-plugin

sudo yum docker-compose-plugin

Step 2 Create a docker-compose.yml file:

The docker-compose file is basically comprised of two services: inaccel-vadd and init and a volume shared among the two services. Init runs first and downloads the right configuration file (bitstream) usinginaccel/cli docker image and finally stores it in the shared volume. inaccel-vadd, that depends on init and uses inaccel/vadd image, is then able to invoke the accelerated vadd application.

Intel PAC A10Xilinx AWS VU9P F1Xilinx U200Xilinx U250Xilinx U280Xilinx U50

38d782e3b6125343b9342433e348ac4c

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/intel/pac_a10/38d782e3b6125343b9342433e348ac4c/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

AWS | shell-v04261818_201920.2

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/aws-vu9p-f1/dynamic-shell/aws/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

xdma_201830.2

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/u200/xdma_201830.2/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

Azure | gen3x16_xdma_shell_2.1

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/u250/gen3x16_xdma_shell_2.1/azure/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

xdma_201830.2

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/u250/xdma_201830.2/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

xdma_201920.3

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/u280/xdma_201920.3/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

gen3x16_xdma_201920.3

docker-compose.yml

services:
  inaccel-vadd:
    depends_on:
      init:
        condition: service_completed_successfully
    image: inaccel/vadd
    volumes:
    - volume:/var/lib/inaccel
  init:
    command:
    - bitstream
    - install
    - https://store.inaccel.com/artifactory/bitstreams/xilinx/u50/gen3x16_xdma_201920.3/vector/1/1addition
    image: inaccel/cli
    volumes:
    - volume:/var/lib/inaccel
volumes:
  volume:
    driver: inaccel

Step 3 Run vector addition example:

docker compose run inaccel-vadd

Step 4 Clean environment:

Delete any containers or volumes created as well as docker-compose.yml file itself:

docker compose down --volumes
rm docker-compose.yml

Step 1 Create a pod.yml file:

To deploy FPGA accelerated applications to Kubernetes, we have to first of all enable InAccel FPGA Operator. To do so, we use the inaccel/fpga: enabled label. Apart from that, to target particular FPGA types, we add a nodeSelector to our workload specification (e.g. intel/pac_a10: 38d782e3b6125343b9342433e348ac4c). Additionally, we specify a resource limit to configure workloads to consume FPGAs (e.g. intel/pac_a10: 1). Finally, we are able to pre-fetch any required bitstreams, using a simple annotation and the bitstreams' URL.

Intel PAC A10Xilinx AWS VU9P F1Xilinx U200Xilinx U250Xilinx U280Xilinx U50

38d782e3b6125343b9342433e348ac4c

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/intel/pac_a10/38d782e3b6125343b9342433e348ac4c/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        intel/pac_a10: 1
  nodeSelector:
    intel/pac_a10: 38d782e3b6125343b9342433e348ac4c
  restartPolicy: Never

AWS | shell-v04261818_201920.2

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/aws-vu9p-f1/dynamic-shell/aws/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/aws-vu9p-f1: 1
  nodeSelector:
    xilinx/aws-vu9p-f1: dynamic-shell
  restartPolicy: Never

xdma_201830.2

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/u200/xdma_201830.2/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/u200: 1
  nodeSelector:
    xilinx/u200: xdma_201830.2
  restartPolicy: Never

Azure | gen3x16_xdma_shell_2.1

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/u250/gen3x16_xdma_shell_2.1/azure/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/u250: 1
  nodeSelector:
    xilinx/u250: gen3x16_xdma_shell_2.1
  restartPolicy: Never

xdma_201830.2

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/u250/xdma_201830.2/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/u250: 1
  nodeSelector:
    xilinx/u250: xdma_201830.2
  restartPolicy: Never

xdma_201920.3

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/u280/xdma_201920.3/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/u280: 1
  nodeSelector:
    xilinx/u280: xdma_201920.3
  restartPolicy: Never

gen3x16_xdma_201920.3

pod.yml

apiVersion: v1
kind: Pod
metadata:
  annotations:
    inaccel/cli: |
      bitstream install https://store.inaccel.com/artifactory/bitstreams/xilinx/u50/gen3x16_xdma_201920.3/vector/1/1addition
  labels:
    inaccel/fpga: enabled
  name: inaccel-vadd
spec:
  containers:
  - image: inaccel/vadd
    name: inaccel-vadd
    resources:
      limits:
        xilinx/u50: 1
  nodeSelector:
    xilinx/u50: gen3x16_xdma_201920.3
  restartPolicy: Never

Step 2 Deploy pod.yml file and inspect the logs:

kubectl apply --filename pod.yml
kubectl wait --for condition=ready --timeout -1s pod/inaccel-vadd
kubectl logs --follow pod/inaccel-vadd

Step 3 Delete the pod created for the accelerated vadd application:

kubectl delete pod/inaccel-vadd

Recap-

In this short section, we learned how we can deploy a simple pre-compiled application targeting datacenter FPGAs through Docker or Kubernetes.
Next, we are going to package our own accelerator and write an application from scratch to cover all the scenarios of a cloud deployment targeting FPGA resources.

Last update: May 15, 2022