Workbench on Kubernetes - Introduction
Instance Types
The k8s sessions run on specific instance types which are preselected on the admin side. The instance type selection matters to ensure that Workbench is not limited by instance type capabilities. For example, when using elastic file system (EFS) on AWS, it is important to make use of network-optimized instances to best support EFS throughput capabilities.
In general, memory-optimized instances are preferred as their CPU/Memory ratio favors memory, which is usually the more important resource for Workbench sessions. The following instance types are currently available to the autoscaler:
| Instance Type | CPU | Memory | Costs per hour |
|---|---|---|---|
r5n.large |
2 | 16 | $0.036 - 0.037 |
r5n.xlarge |
4 | 32 | $0.074 - $0.099 |
r5n.2xlarge |
8 | 64 | $0.2 - $0.4 |
r5n.4xlarge |
16 | 128 | $0.36 - $0.95 |
| ———— | – | —– | ————– |
r6i.large |
2 | 16 | $0.0383 - $0.0390 |
r6i.xlarge |
4 | 32 | $0.0811 - $0.105 |
r6i.2xlarge |
8 | 64 | $0.01624 - $0.3526 |
r6i.4xlarge |
16 | 128 | $0.336 - $0.3747 |
More instance types or families can be added if needed.
Resource Requests
The default session resources aim to represent a reasonable default for new sessions. Depending on the need of your current task, you might want to increase the respective values.
CPU resources are often shared between multiple users on a machine (as the CPU is idle most of the time anyway). Memory requests are blocked for the individual users to prevent memory overflow of the underlying machine.