This guide offers recommended configurations and settings unique to Azure Kubernetes Service (AKS). These should be used to augment the generic requirements offered on our primary requirements page.
Instance types
- Minimum:
Standard_D8s_v4
- Recommended:
Standard_D16s_v4
Storage
Unfortunately, we have found that Azure’s built-in, managed Network File System (NFS) service, Azure Files NFS, does not provide an acceptable performance level for use with Data Science & AI Workbench. We have not yet had the opportunity to evaluate Azure NetApp Files.
For this reason, Anaconda recommends creating a separate Virtual Machine for hosting NFS storage. Follow the recommendations offered in this document, with these Azure-specific recommendations:
- The
Standard_D4s_v3
machine type is suitable for this purpose. - Azure tightly couples disk size and IOPS performance. Anaconda recommends a minimum disk size of 1 TiB to ensure good performance.
This server can be the administration server as well.
Network
Azure offers two different networking options for AKS clusters. Both approaches are compatible with Workbench, so the determination depends upon your larger networking needs.
AKS uses a LoadBalancer which by default sets TCP idle timeouts to 4 minutes, and enables TCP resets. This may affect any user workload that depends on a continuous TCP connection.
GPUs
Please see this Azure guide for adding GPU resources to your AKS cluster.
Was this page helpful?