How Fuzzball's HPC runs on Kubernetes (k8s)
CIQ HPC engineers Brian Phan and Jonathon Anderson explain in one minute how Fuzzball uses Kubernetes without the overhead of earlier containerized HPC attempts. Fuzzball Orchestrate runs as microservices on Rancher Kubernetes Engine and manages workflows plus the data and software images they need. Jobs are submitted to Orchestrate, which stages containers and data onto compute nodes running Fuzzball Substrate, a task runner that executes the work on bare metal.
Workflows are collections of jobs related by dependencies. Orchestrate processes the workflow, but every compute job and ancillary task such as pulling an image or handling data ingress and egress runs outside Kubernetes on the Substrate resources.
Key takeaways
- Fuzzball Orchestrate runs as microservices on Rancher Kubernetes Engine to manage workflows, data, and software images.
- Fuzzball Substrate acts as a task runner on compute nodes, receiving staged containers and data from Orchestrate.
- Compute jobs and ancillary tasks like image pulls and data ingress or egress execute outside Kubernetes on bare metal.
Questions this video answers
What runs on Kubernetes in Fuzzball and what does not?
Only Fuzzball Orchestrate, the set of microservices that manages workflows and stages data and software images, runs on Kubernetes. Jobs are executed by Fuzzball Substrate on your compute nodes, along with ancillary work such as pulling images and moving data, so workloads get bare-metal performance.
About this video
Fuzzball is an HPC workflow orchestration platform that runs on Kubernetes. It's designed so you get bare-metal performance without any of the overhead that's come with previous attempts at containerized HPC.
Learn more from Brian Phan & Jonathon Anderson, CIQ's HPC engineers.
This video is part of the Fuzzball playlist. Browse every CIQ video by product and topic.
Transcript
we're using kubernetes to run the various microservices to manage the workflows and data software images that are being orchestrated for your particular workflow orchestrate basically will run the job on bare metal orchestrating the workflow using these microservices fuzzball orchestrate is running on rer kubernetes engine jobs are submitted into fuzzball orchestrate on your compute nodes we have something called fuzzball substrate you kind of think of this as a task Runner that runs various jobs that are submitted to orchest so orchestrate will stage the various data and software containers onto the substrate node and running your job on this particular compute mode fuzzball executes workflows those workflows
are collection of jobs that can be related to each other through dependencies the workflows are processed in orchestrate compute jobs that you want to run or ancillary jobs like pulling an image or doing Ingress erress those all execute outside of kubernetes on your compute resources that are running substr
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.
