Question about karmada-scheduler-estimator component performance in large-scale clusters scenarios

[zhy76]

Please provide an in-depth description of the question you have:
If we want scheduling to be as precise as possible to avoid situations where a workload cannot be launched even though there are resource budgets available for developers' side, I see karmada has the karmada-scheduler-estimator component which can accurately calculate the number of schedulable replicas in each cluster. I wonder the performance of this component in large-scale clusters. This component seems to be centrally deployed in the host cluster and needs to monitor and calculate resources for all pods across all clusters. I wonder why not adopt a method where the calculation is done in the member clusters, and the results are aggregated in the host cluster, this way might reduce the pressure on the host cluster?
What do you think about this question?:
I would also like to know what are the best practices for scenarios requiring accurate calculation of available resources in clusters for more precise scheduling? How do mainstream approaches trade-off performance and scheduling accuracy?Thanks~
Environment:

• Karmada version:
• Kubernetes version:
• Others: