How Fuzzball Integrates with Your Existing HPC Infrastructure with No Rip and Replace
David Godlove explains how Fuzzball fits into hardware and tooling an organization already runs. Provisioner keywords decide where workflows execute: static for nodes Fuzzball manages itself, AWS for EC2 instances, and Slurm or PBS to schedule onto an existing cluster while users keep submitting jobs the way they always have. Several provisioners can coexist, combining dedicated nodes with cloud bursting and legacy schedulers.
Policy expressions add rules such as limiting GPUs to an AI lab, setting conditions for AWS spot instances, or directing large-memory jobs to a specific partition. A CIQ plugin runs Nextflow pipelines directly on Fuzzball, and integrations cover OCI image registries, object stores, MPI, GPUs and high-speed fabrics, alongside Rocky Linux, Warewulf Pro and Ascender Pro.
Key takeaways
- Provisioner keywords tell Fuzzball how to obtain resources: static for its own nodes, AWS for EC2 instances, Slurm or PBS for existing clusters.
- Using Slurm or PBS as a provisioner adds container orchestration to an existing cluster without disrupting users' normal job submission.
- Multiple provisioners can run in one deployment, mixing dedicated nodes, cloud bursting to AWS and integration with legacy schedulers.
- Policy expressions restrict who can use GPUs, when spot instances may be allocated, and which partition large-memory jobs must request.
- A CIQ plugin executes Nextflow pipelines directly on Fuzzball, much like pointing Nextflow at an AWS or Slurm backend.
Questions this video answers
Does Fuzzball replace Slurm or PBS?
No. Fuzzball can use Slurm or PBS as provisioners, asking those batch schedulers to allocate resources for users' jobs. That adds modern container orchestration to an existing HPC cluster while preserving the ability to submit Slurm and PBS jobs directly, and the same deployment can also provision static nodes or AWS EC2 instances.
How does Fuzzball control who can use GPUs or cloud resources?
Policy expressions define rules about who may use computational resources and under what circumstances. The examples shown restrict GPU resources to members of an AI lab, specify the conditions under which users can allocate AWS spot instances, and require large-memory jobs to be requested through a designated partition.
About this video
Already running Slurm or PBS? Have GPU resources locked to specific teams? Bursting to AWS? Fuzzball doesn't ask you to start over — it integrates with the infrastructure you already have. In this video, CIQ's David Godlove walks through Fuzzball's flexible provisioner and policy system, showing how IT administrators and AI/ML engineers can:
Mix and match provisioners — static nodes, AWS EC2 spot instances, Slurm, and PBS — in a single Fuzzball deployment Use policy expressions to control who can access GPU resources, spot instances, and high-memory partitions Add modern container orchestration capabilities to existing HPC clusters without disrupting existing Slurm or PBS job submission workflows Execute Nextflow pipelines natively on Fuzzball using the CIQ plugin Integrate with OCI image registries, object stores, MPI, GPUs, and high-speed fabrics
Fuzzball is also part of CIQ's complete infrastructure stack — built to work alongside Rocky Linux, Warewulf Pro, and Ascender Pro — giving you the flexibility to mix and match based on your environment's specific needs.
This video is part of the Fuzzball playlist. Browse every CIQ video by product and topic.
Transcript
Let's talk about integration. Fuzzball gives you the power and the flexibility to integrate with your existing hardware and tooling. For instance, you can choose how Fuzzball will provision computational resources for your users workflows. The static provisioner keyword tells Fuzzball that you want it to provision resources for your users. The AWS provisioner keyword lets Fuzzball know that you should call out to AWS to request EC2 instances for your users jobs. need to integrate with an existing PBS or slurm cluster. Fuzzball can use these batch scheduling systems to provision resources too. This allows you to add modern container orchestration capabilities to your existing HPC cluster while preserving your users's ability to submit slurm and PBS jobs.
The slurm and PBS keywords tell Fuzzball to use the respective batch scheduling systems to provision resources on behalf of your users to run their computational jobs. You can even use multiple provisioners in a single setup allowing fuzzball to provision dedicated nodes itself. Um and then use AWS for cloud bursting and uh use slurmur or PBS for existing HPC clusters that you want to integrate with. Policy expressions allow you to create rules determining who can use your computational resources and under what circumstances. For instance, here we have a rule specifying that only users of an AI lab are allowed to use GPU resources. We have another rule specifying under what conditions users can allocate AWS spot instances.
And a final rule saying that users who need large memory jobs should request them using this learn partition. Need to run Nexflow pipelines. CIQ has a plugin allowing you to execute Nexflow pipelines directly on Fuzzball the same way that you would point them to an AWS or a slurm backend. Fuzzball also integrates seamlessly with tools like OCI image u registries object stores as well as HPC specific tools like MPI GPUs and high-speed fabrics. Fuzzball is also part of a complete infrastructure stack starting with Rocky Linux serving up the operating system, Warewulf Pro for cluster provisioning and ascender pro for automation. All of these are designed to work together with flexibility, allowing you to mix and match based on your specific needs.
Built for scale. Chosen by the world’s best.
2.75M+
Rocky Linux instances
Being used world wide
90%
Of fortune 100 companies
Use CIQ supported technologies
250k
Avg. monthly downloads
Rocky Linux
9
Enterprise products
Spanning the kernel to the orchestrator
Have questions about your infrastructure?
Talk to a CIQ engineer about Rocky Linux, HPC, and AI infrastructure.