Accelerator-K8S

Accelerator-K8S is a device plugin works with Accelerator-Docker

Getting Started

Dependencies

Kubernetes
Accelerator-Docker

Installation

Deploy as a daemonset (Recommended)

kubectl create -f https://raw.githubusercontent.com/KAIST-NCL/Accelerator-K8S/master/acc-k8s.yml

Install manually by cloning and building

$ git clone https://github.com/KAIST-NCL/Accelerator-K8S.git
$ cd Accelerator-K8S
$ make
$ sudo make install

$ out/acc-k8s

for every node

How to Use

Prerequisites

First, you need to make sure that you installed Accelerator-Docker properly.

default-runtime of docker daemon should be set to acc-runtime. Check /etc/docker/daemon.json
In Accelerator-Docker setting(/etc/accelerator-docker/device.pbtxt), accelerator type should be named following kubernetes resourece naming rule. Refer to this.

Enabling Accelerators in Containers

You can require your accelerators as a resource like below:

apiVersion: v1
kind: Pod
metadata:
  name: demo-pod
spec:
  containers:
    - name: demo-container
      image: ubuntu:16.04
      resources:
        limits:
          {ACCELERATOR_TYPE}: 1

'Accelerator type' is configured in Accelerator-Docker configuration file and you can check it by using acc-manager list command

Example)

$ acc-manager list
+------------+--------------+------------------------+----------------+----------------+----------+
|     ID     |     Name     |          Type          |    PCI-Slot    |     Status     |  Holder  |
+------------+--------------+------------------------+----------------+----------------+----------+
| 234750     | QuadroM2000  | nvidia.com/gpu         | 0000:02:00.0   | Available      | 0        |
| 1121730    | KCU-1500     | xilinx.fpga/kcu-1500   | 0000:01:00.0   | Available      | 0        |
+------------+--------------+------------------------+----------------+----------------+----------+

If you want to use KCU-1500 for your pod, you can set container limit as follows.

  containers:
    - name: demo-container
      image: ubuntu:16.04
      resources:
        limits:
          xilinx.fpga/kcu-1500: 1

Authors

KAIST NCL

Sunghyun Kim

License

This project is released under the BSD-3-Clause, see LICENSE for more information.

Acknowledgement

This work was supported by 'The Cross-Ministry Giga KOREA Project' grant funded by the Korea government (MSIT) (No.GK19P0400, Development of Mobile Edge Computing Platform Technology for URLLC Services)