Last Updated: 1/13/2020
For Azure Marketplace, see original Microsoft Azure announcement
AWS Marketplace makes it easy to launch Graphistry in any GPU-capable AWS region. This tutorial shares a visual guide of launching Graphistry, trying your first notebook, and help with common configuration options and occasional error handling. When you’re ready, head to AWS Marketplace! And, if you haven’t heard about Graphistry or Graphistry’s AWS launch, learn about why we did it and what’s new.
Launch a private Graphistry instance from AWS Marketplace
All you need to start is an AWS account with GPU quota. The following instructions walk through using AWS Marketplace to subscribe to Graphistry (free), launch a Graphistry AMI on a GPU instance type of your choice (regular instance fees) as part of your regular or new VPC + security groups, and start using it. Everything is encapsulated inside your VM, and you can see what is included in our architecture diagrams. For a sample Neptune DB integration scenario, you can peek into the architecture on the AWS blog. The process can take as fast as 6 minutes based on your AWS VPC/subnet setup. If you are not familiar with launching AWS AMIs or suspect VPC/subnet security issues, our team is happy to jump on a call or answer any questions.
1. Subscribe Screen
- If prompted, accept terms and wait through ‘Pending’ period
- Do not configure the annual contract discounted prepay option: Total contract price should be $0
- Pricing occurs later when you pick the type of GPU, e.g., the g4dn.xlarge Individual/Startup configuration vs. larger Team-level servers
- Continue to Configuration
2. Configure Screen
- Version: Default – the latest Graphistry version
- Region: Pick a GPU-capable region close to your users or data
- Tip: Teams – Search for G4DN (Nvidia T4), P3 series (Nvidia V100), or P4 series (Nvidia A100) on the EC2 instance comparison chart, with US East and Oregon as great options in the US.
- Choose 1 GPU / 4 vCPUs for a discounted individual (g4dn.xlarge) server
- Choose 1+ GPUs and 8+ CPUs for running concurrent users (see additional guidance)
Tip: If your AWS account does not have GPU quota, you can request it to get added.
Tip: We recommend starting with pay-as-you-go: You will only be billed for Graphistry software + AWS utilization while the server is running, and only for storage whenever you stop the instance. When your usage pattern clarifies, you can pick appropriately discounted tiers by using reservations, or reaching out to our team for additional options.
- Do not configure the annual contract discounted prepay option: Total contract price should be $0
- Tip: Teams – Search for G4DN (Nvidia T4), P3 series (Nvidia V100), or P4 series (Nvidia A100) on the EC2 instance comparison chart, with US East and Oregon as great options in the US.
- Continue to Launch
3. Launch Screen
- Action: Default – Launch from Website
- Instance: Default is recommend – g4dn family are the smallest Nvidia-RAPIDS-compatible GPU, with 8+ core recommended
- Security group: We recommend ‘Create new based on Seller Settings’ – 22 (SSH for admins), 80 (initial web port), and 443 (automatic/custom TLS once you assign a domain). In secure deployments, when ready, you likely want to disable 22 and 80.
- Key pair: Use existing if you have a copy of the files, else create new (and save), and likely `chmod 400 myfile.pem` (or chmod 644 myfile.pem)
- Launch
- Wait till ‘Status Checks: 2/2 checks’
- Note the `i-YourInstanceID` value, which you will use as a password
- Follow the public IP / DNS link from your EC2 console to start your first Graphistry session!
Login & launch your first visualization & notebook
1. Login
- In a browser, go to the public DNS provided by the EC2 console
- Upon starting the system, you need to wait ~5 minutes, and until then, may see 5XX errors or otherwise incomplete page loads
- AWS Marketplace images create the initial admin with unguessable credentials `admin` / `i-YourAWSInstanceID`. Upon login, we recommend changing the password, and to aid resets, email address, and then save them in a password manager.
NOTE: Non-AWS-Marketplace users will create the initial web admin account on first load, after which the registration system auto-closes.
2. Open a notebook
- ‘Analyst Getting Started’ -> ‘See Notebook’
- In sequence, run each code cell by clicking on it and entering ‘shift-enter’
- On cell ‘g.plot()’, on the upper right of the graph where it says ‘workbook’, click the left-most icon for ‘Open in a new window’, and click the Graphistry logo to begin
- Follow the UI guide on how to interact with the notebook
3. Try more notebooks
- Try other notebooks: for developers, CSV upload of your own data, various DBs, …
- Make a copy of a notebook you want to use as a template
- You can always edit any notebook cell, re-run it, and save your changes/outputs
- Best practice is to make sure you can ‘run-all’ in a notebook and it gives the full results
Recommended additional config & troubleshooting:
See the Graphistry AWS Marketplace FAQ:
- Errors: Handle common AWS issues like no available GPUs
- IP/Domain: Set an elastic IP or domain to have a stable URL across restarts
- Enable Graphistry notebooks to communicate with secure databases
- If `pip install` does not have your package, SSH into the instance, and then the Docker container as root
- Make sure the Graphistry IP is permitted by the security group for your DB
- Disable ports 22/80, and for server (non-web) admins who need direct server access, create limited privilege Ubuntu accounts
- Create accounts for your users
- As an admin, consider creating a non-admin web account for yourself
- Instead of sharing root server accounts with admins, consider only sharing web admin access with them
- Likewise, when giving server access, sharing restricted individual server accounts (with Docker service rights) is preferred
- Setup TLS, disable 22/80 when not administrating, and restrict server access to your VPN
Next steps & further reading