GCP

Google Cloud Platform has a free-tier, with free credits that are easy to qualify for.

Login to Google account

You’ll need a personal Google account to use the Google Cloud Platform. Do not use your umich email address. The following steps are adapted from Google Cloud’s Quickstart using a Linux VM, though they have been adapted to the course.

Create project

Go to the Google Cloud Platform Project Selector page and create a project. Click AGREE AND CONTINUE if you agreed to Google’s Terms of Service.

Create a project by clicking on the Create Project button.

Give your project a name unique to you and click Create.

Add billing

After selecting your newly created project, the side menu may have additional items pre-pinned, however the two items we need Billing and Compute Engine are easily identifiable. Please consult the teaching staff if you couldn’t find any menu items referenced in this spec.

Click billing to add a billing method to your project.

When you fill out your billing information, select “individual” as account type. Make sure you see something like this:

Add a credit or debit card. If your back end qualifies for free-tier (it should), this card will not be charged. Select START MY FREE TRIAL. Return to the project console.

Enable Compute Engine API

Visit GCP’s Compute Engine API site

and select ENABLE.

Prepare SSH keys

Before creating the VM instance, we’ll set up SSH keys for secure access. The back-end specs in this course will assume that you have a user with username “ubuntu” created on your instance. The specs further assume that you’re doing all your back-end work, building your server, under user “ubuntu”. You’re free to build under a different username, e.g., your Google account name, however you will then have to map between the instructions in the specs and your setup. More importantly, we will NOT be able to help you should you need help debugging your back-end server. To build your back end as user ubuntu, please do the following:

  laptop$ sudo vi /etc/wsl.conf

  [automount]
  options = "metadata"

  #                       👇👇👇👇👇👇👇👇👇👇👇👇👇
  laptop$ ls /mnt/c/Users/YOUR_WINDOWS_USERNAME/

First generate a public/private key pair for user “ubuntu” in a safe place you can easily remember, for example YOUR*LABS*FOLDER.:

#         👇👇👇👇👇👇👇👇👇👇
laptop$ cd YOUR*LABS*FOLDER
laptop$ ssh-keygen -C ubuntu

when ssh-keygen prompts you for the file in which to save the key, enter “reactive.pem”. It will then prompt you for a passphrase. Leave it empty. Hit return or enter, twice. Your identification (private key) would have been saved in reactive.pem and your public key in reactive.pem.pub. You can view the content of your public key for posting to Google below by:

laptop$ cat reactive.pem.pub

Add SSH keys to project metadata

Go to GCP Metadata page

Open the SSH KEYS tab and click EDIT.

On the edit page, click + ADD ITEM, copy and paste the content of your “reactive.pub” to the empty box that + ADD ITEM brought up, and hit the blue SAVE button. The content should be something that start with ssh-ed and end with your username (e.g. ubuntu).

Your SSH KEYS tab should now list “ubuntu” under Username with its corresponding public key:

Configure firewall rules

Next set up the firewall to allow us to test the web server we’ll be setting up later.

Go to VPC network Firewall page and click the Create firewall rule.

On the Create a firewall rule dialog box, scroll down to “Targets”. Enter http-server into the “Target tags” field. Enter 0.0.0.0/0 into the “Source IPv4 ranges” field. Check TCP and enter 8000 into the “Ports” field.

Click CREATE.

Create VM instance

Now we’re ready to create the VM instance. Return to the console. Navigate to Compute Engine > VM Instances.

Select CREATE INSTANCE.

Choose instance

As discussed in the “Hosting cost” section of the spec, for this course, we assume you are running a free-tier backend and will provide only instructions to set up a free-tier backend, which can only run the tinyllama LLM model. If you want to run other LLM models, choose the VM instance with enough resources to host your model of choice. You would have to pay for VM instances with more resources.

Review the free tier options at the Google Cloud Free Program page by scrolling to the section titled “Free Tier usage limits”. Look under the “Compute Engine” section and check regions eligible for free tier. Free tier usage is governed by time used. Currently, an e2-micro in Oregon, Iowa, or South Carolina is eligible for free tier if used for the number of hours in a month.

Give the instance a name and carefully select the regions that are available as free tier with an e2-micro configuration.

Configure boot disk

On the left menu, click on OS and storage, and click the Change button. On the Boot disk page, in the “Operating system” field choose Ubuntu. In the “Version” field, choose Ubuntu 24.04 LTS for x86/64. In “Boot disk type”, be sure to select Standard persistent disk. Any other Boot disk type option will be extra charge!. For the “Size” field, students have reported that 10GB is not enough to install Ollama. GCP free tier allows up to 30 GB free storage per month. However, this 30GB is across all your instances, and you cannot shrink a volume once created, you can only expand it. If you already have a free GCP instance in another course or project, you will likely be billed if your total is above 30 GB.

Set your storage to 30 GB:

Click SELECT.

Configure firewall and networking

On the left menu, below OS and storage, click Networking. In the Firewall section, check both Allow HTTP traffic and Allow HTTPS traffic.

Startup script

At the bottom of the left menu, click Advanced. Navigate to the Automation section and find Startup script.

Set up a 4GB swapfile so that your server can survive load spikes. Paste the following script:

#! /usr/bin/env bash

fallocate -l 4G /swapfile  # free -h to see allocated size
chmod 600 /swapfile
mkswap /swapfile
swapon /swapfile

Create the instance

Review all your settings:

• Name and region
• Machine type: e2-micro
• Boot disk: Ubuntu 24.04 LTS, Standard persistent disk, 30GB
• Firewall: HTTP and HTTPS allowed
• Network tags: http-server
• Startup script added

Finally, click Create at the very bottom of the page to create your VM instance. Wait for the instance to initialize.

When the loading animations are done, write down the external IP address shown on the screen. In the remainder of this spec, and actually, in all other tutorials of the course, we will refer to your “external IP” as YOUR_SERVER_IP. You’ll never need the internal IP (and GCP doesn’t provide any public DNS for your instance).

ssh to instance

To ssh to your GCP instance as user “ubuntu” using your private key, reactive.pem, you must first set its permissions to read-only. In the following, YOUR_SERVER_IP always refers to the external IP address you’ve noted down earlier.

laptop$ chmod 400 reactive.pem
laptop$ ssh -i reactive.pem ubuntu@YOUR_SERVER_IP
#                                 👆👆👆👆👆👆👆👆👆

PS laptop> icacls reactive.pem /grant "$($env:username):(r)" /inheritance:r
PS laptop> ssh -i reactive.pem ubuntu@YOUR_SERVER_IP
#                                    👆👆👆👆👆👆👆👆👆 

Persistent swapfile

To mount the 4G swapfile we created above persistently after each reboot, do:

server$ sudo su
server# echo '/swapfile none swap sw 0 0' >> /etc/fstab
server# exit

Increasing the disk size

If you have set up your instance with less than the 30 GB suggested above and got an error message that you don’t have enough storage, to increase your storage, in Google Cloud Console, go to the Disks page, then click the name of the disk that you want to modify. Note that there’s an upper limit of 30 GB of total free-tier storage across all your GCP instances.

On the details page, click “Edit”, then enter a larger size, e.g., 20 GB, for the “Size” and click “Save” at the bottom. You may need to wait a few minutes for the modification to take effect.

Once the volume size modified, you also need to resize the partition and filesystem inside the VM: ssh to your sever, then check your storage devices (disk, partitions, and logical volumes).

server$ lsblk

From the list of block devices, look for /dev/sda. This is your disk storage and it should show the modified size (20GB) of your volume. Your root partition would be listed as /dev/sda1. Expand your root partition (/dev/sda1) to use all of the free space on the volume (/dev/sdda):

server$ sudo growpart /dev/sda 1
server$ sudo resize2fs /dev/sda1  

The above commands grow the root partition and resize the filesystem.

You can verify the updated partition size by:

server$ df -h

You should see /dev/root mounted at / and that its usable space is as you’ve set.

Stop instance