Conceptual
Browse Instruct Curate
Login

Reading Memory and CPU Metrics to Tell Starved From Slow

The outcome requires a run that finishes and costs roughly what was predicted, and the commonest failure is renting the wrong size of machine. Without this node the learner's only response to 'slow' is to rent something bigger, which is the mechanism by which grant money is burned. It also feeds directly into the decision of whether to stop a run.