A company runs a provisioned Amazon EMR cluster using general-purpose Amazon EC2 instances. EMR managed scaling is configured to vary the task node count from one to five for a daily, long-running Apache Spark ETL job. Each run quickly scales the cluster to five task nodes; CPU utilization often hits its limit while memory stays below 30%. The company wants to change the EMR configuration to lower the cost of running the daily ETL job. Which option will most cost-effectively meet this requirement?

The correct option is to switch the task node type to compute-optimized EC2 instances. The problem states that CPU utilization often hits its limit while memory stays below 30%. This indicates a CPU-bound workload. Compute-optimized instances (e.g., C-series) provide a higher ratio of CPU to memory and are more cost-effective for such workloads compared to general-purpose instances. Increasing the maximum number of task nodes would increase cost without addressing the underlying resource imbalance. Changing to memory-optimized instances would worsen the problem by providing more memory, which is already underutilized, and less CPU. Reducing the scaling cooldown period would affect how quickly the cluster scales but wouldn't change the fundamental resource allocation per node or address the CPU bottleneck.