We currently set the default memory usage for time-series collections at a conservative 2.5%, which can lead to performance issues in which a user mitigates by scaling up faster and support requests to raise this limit. Additionally, since this parameter isn't documented, users are often unaware of its potential performance benefits, especially for time-series workloads which can cause churn for self service users due to misconfigurations which aren't optimal and the perception of poor performance or lack of tunability.

Based on our experience raising these limits in many user deployments since 5.0, we can safely increase the default memory usage. The proposal is to raise the default to 5%, which will improve performance and reduce support needs for efficient time-series deployments, specifically for:

• timeseriesSideBucketCatalogMemoryUsageThreshold
• timeseriesIdleBucketExpiryMemoryUsageThreshold