I got some feedback from a customer today that they'd like more granular control over the cost/performance trade-off:
Shutting down resources is great for saving costs, but unfortunately some resources (especially e.g. custom notebook kernels with large images) may take several minutes to restart when a user wants to access them again - which can be frustrating & impact productivity.
To balance user convenience with operational cost, the optimal trade-off is very different for a slow-to-start custom kernel on an ml.t3.medium than, say, a Data Wrangler session on an ml.m5.4xlarge at (approximately) 18x the cost per instance-hour.
It would be great if this extension could offer configuration to shut down higher-cost resources faster than cheap ones!
I got some feedback from a customer today that they'd like more granular control over the cost/performance trade-off:
Shutting down resources is great for saving costs, but unfortunately some resources (especially e.g. custom notebook kernels with large images) may take several minutes to restart when a user wants to access them again - which can be frustrating & impact productivity.
To balance user convenience with operational cost, the optimal trade-off is very different for a slow-to-start custom kernel on an
ml.t3.mediumthan, say, a Data Wrangler session on anml.m5.4xlargeat (approximately) 18x the cost per instance-hour.It would be great if this extension could offer configuration to shut down higher-cost resources faster than cheap ones!