Load-Based Sizing

Load-based sizing automatically adjusts Yeedu cluster resources after jobs are submitted, according to the workload actually demanded of them. The distinction matters: we react to submitted work rather than attempting to anticipate it, so the cluster's size follows the queue instead of a prediction that might not materialise.

That's a deliberate trade. You won't get capacity provisioned in advance of a burst, and in exchange you don't pay for capacity a forecast turned out to be wrong about.

How It Works

• When a job is submitted, Yeedu evaluates the current cluster load.
• Based on the number of running and queued jobs, compute resources are adjusted.
• If job load increases → the cluster scales up to Max Instances
• If job load decreases → the cluster scales down to Min Instances

Key Parameters

• Min / Max Instances
• Parallel job execution limits
• Instance type (Compute)

This ensures the cluster scales only in response to submitted jobs, providing optimal performance while controlling compute costs.