Adrian Torrie1
1XENON Systems, Springvale, Australia
Biography:
Adrian is a skilled solutions architect with over 17 years of experience building and maintaining data systems from the ground up.
He has broad skills across all IT domains, with specific depth in automating data analytics/machine learning/artificial intelligence solutions in hybrid environments, along with the necessary skills for training others in this specialty.
Adrian's last role had the primary focus of enabling self-service, via Platform Engineering, for Data Scientists and Data Engineers, allowing rapid iteration of models and their release.
He excels in delivering secure solutions for large scale, high performance computational requirements using modern automation approaches.
Abstract:
Containerising scientific workflows is a revolution, but traditional HPC users worry: "Can Kubernetes handle my complex workloads with familiar scheduling?". This session unveils the power of Kubernetes as a HPC platform that bridges the gap for HPC users.
Slurm/PBS-like Scheduling: Manage HPC workloads with hierarchical queues, gang scheduling, and familiar job priorities, using tools such as Yunikorn – a smoother transition from your existing platform.
Lift-and-Shift: Migrate monolithic applications, VMs, and MPI jobs onto Kubernetes with minimal disruption. Leverage tools like virtual clusters (vcluster), KubeVirt, and KubeApps for a streamlined process.
This session is for you if:
– You are an HPC user wrestling with scheduling complexities in Kubernetes.
– You are looking to migrate existing HPC environments (applications, VMs, MPI jobs) to Kubernetes.
– You want to leverage familiar scheduling concepts with Kubernetes for seamless workflow integration.