Quickly Deploy IBM Cloud Private with GPU Capability

IBM Cloud Private is an IBM-supported wrapper around Kubernetes. This tutorial will get you up and running with ICp in a few short steps. While the instructions do assume you want to use GPU enabled systems, you can use the same instructions to run the product on any cloud infrastructure.

You will first need a Linux or Mac computer (or VM) to drive this lab and host Cloudsoft AMP. It should have at least 20GB of disk space, one CPU core, and 2GB of RAM. You will need to create a ssh key (if you don’t already have one) using ssh-keygen.

Download AMP (tarball format) from cloudsoft.io.

If using a VM, create a user to own AMP and create a ssh key:

sudo adduser amp
su — amp
ssh-keygen # accept defaults here, no passphrase

Alternatively, if you are using your own computer, you can use your default user and ssh key.

Download the NVIDIA CUDA installer into the home directory of the AMP user (or your default user) on your AMP server. Use this URL to get the proper version:

Or use the direct link:

wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64-deb

This tutorial is set up to use cuda-repo-ubuntu1604–9–0-local_9.0.176–1_amd64-deb. The yaml blueprint below can be updated to use different versions.

On your VM (or local Linux or Mac computer) unpack the tarball and start AMP. It is highly recommended that you edit brooklyn.cfg and change the default UI Console password.

tar xvzf Downloads/cloudsoft-amp-karaf-4.latest.tar.gz
cd cloudsoft-amp-karaf-4.8.0
vim etc/brooklyn.cfg # Change the password using your favorite editor
./bin/start amp

Navigate to the AMP console and login as `admin` with password `password`:

http://IP_ADDRESS_OF_YOUR_HOST:8081/

Add a SoftLayer Bare Metal GPU Hourly “location” to AMP using the “Blueprint importer” icon on the AMP console. Copy and paste the YAML from the link below (bare-metal-location.bom) into the importer. Change “your-sl-identity” and “your-sl-api-key” to your SoftLayer API credentials, then click the “Import” button.

AMP Blueprint importer icon

https://raw.githubusercontent.com/rbraddes/BrooklynBlueprints/master/bare-metal-location.bom

This blueprint should work with any Cloud location in AMP. It has been tested with Bare Metal GPU servers in SoftLayer (as defined above) and Virtual Machines (also in SoftLayer). However, it should be cloud agnostic. You can use the “Location manager” in AMP to add any cloud you wish.

Delete all text from the Blueprint importer and then copy and paste the YAML from the link below into the form and re-click the “Import” button:

https://raw.githubusercontent.com/rbraddes/BrooklynBlueprints/master/icp_cluster_ce.bom

You will now see “ICp CE Cluster” in the “Quick launch” section of the AMP GUI main page.

ICp CE Cluster icon in the Quick launch field

When you click on the “ICp Cluster” icon, you will see a Deploy window where you can enter an optional name. For a Location, use the one you created above (bmc-sjc04-gpu1) and then select the initial number of worker nodes you want in your cluster.

When you click “Deploy”, AMP will provision the hourly GPU servers in SoftLayer, then install and configure IBM Cloud Private on the servers.

You can follow the progress on the “App inspector” page in AMP (use the icon on the AMP home page).

You can follow along the deployment process using the “Activities” tab for any of the entities being deployed and drilling down into the Activity links.

Once your cluster is ready, you can access it by finding the public IP address of the master node (expand the application on the App inspector page until you see the master node, click on it, then select the “Sensors” tab on the right side of the page) and go to https://YOUR_PUBLIC_IP:8443/ (user ID is *admin*, default password is *admin*).