Deploy Fuzzball On-Prem: Step-by-Step Guide and Webinar

We’re thrilled to share with you how Fuzzball works for on-premises installations. Learn all about its container-based architecture, how to deploy it to your on-premises infrastructure, and how to run some sample jobs in this webinar.

Join Zane Hamilton as he hosts a Fuzzball install demo with CIQ’s High Performance Computing (HPC) engineers Brian Phan and Jonathon Anderson.

An overview of Fuzzball on-prem

Fuzzball is a workflow orchestration platform designed for modern HPC workloads. Using its intuitive graphical user interface or its workflow definition language, you can build workflows for complex tasks like simulations, rendering, and complex scientific computing.

Many of the difficulties of traditional HPC have been streamlined and rethought with Fuzzball, from user management to data ingress and egress. Its API-first design allows total customization, allowing you to integrate Fuzzball workflow orchestration into your existing and future automated processes.

How Fuzzball uses containers for HPC

Fuzzball runs on Kubernetes and is itself container-based. Others have attempted to use Kubernetes for HPC, but found that it introduces non-trivial overhead that affects performance. With HPC, you want to squeeze every ounce of performance from your hardware that you can get. Fuzzball solves this problem with a novel architecture design.

Specifically, Fuzzball uses Kubernetes to run the microservices that manage workflows, data, and software images that are being orchestrated for each job. At the end of the day, your workflows run on bare metal and are managed by the Fuzzball Orchestrate platform via Kubernetes.

How Fuzzball deploys via Kubernetes

Using the Kubernetes Operator pattern, we can deploy the Fuzzball Orchestrate platform onto an existing Kubernetes cluster. The Operator pattern allows for more flexibility, easier upgrades, and a broader range of integrations, which has been helpful for existing users.

Most of our testing has been on Rancher Kubernetes Engine 2 (RKE), but Fuzzball is likely to deploy without issue for any standards-compliant Kubernetes environments. RKE has proven to spin up environments quickly and easily, so we prefer it internally, but we’ve seen reports of successful deployments on other SIF-compliant K8s platforms.

How Fuzzball executes jobs with Orchestrate and Substrate

When jobs are submitted to Fuzzball, the Orchestrate system sends tasks out to compute nodes that run Fuzzball Substrate, which is essentially a task runner for orchestration jobs. Orchestrate will stage the various data and software containers onto the Substrate node, and eventually, it will end up running your jobs on compute nodes.

If you’re coming from a traditional HPC environment running SLURM, you can think of Fuzzball Orchestrate like the slurmctld and slurmdbd, all in one. Substrate can be thought of as the SLURM daemon that runs on your compute node.

Deployment diagram via CIQ Mountain

Fuzzball can be deployed via CIQ’s secure artifact repository service, Mountain. The diagram below demonstrates that deployment workflow and reflects a simple execution architecture that’s used in the webinar install.

Watch the Fuzzball install webinar

There’s way too much to share in a high-level overview of the webinar. We recommend you watch the whole video for an in-depth look at how Fuzzball and its components are deployed. Then stick around until the end to see how jobs are run and get a sense of what’s coming next for Fuzzball.

Some of the key product benefits featured in the webinar:

• Fuzzball supports CLI and web UI interfaces for workflow management
• Jobs can be defined with resource requirements (cores, memory) and policies (like time limits)
• The web UI provides real-time monitoring and logs for running jobs
• Fuzzball integrates with CIQ Mountain for easy access to software and tools required for deployment
• The Fuzzball web UI allows you to interact with a shell inside a running container

If you have questions, feel free to email us at info@ciq.com. Check out the Fuzzball documentation at beta.fuzzball.io/docs. And if you want to give Fuzzball On-Prem a try in-house, just reach out with this form.