As part of the PERF-6051 project, we conducted scalability testing for out-of-cache workloads. We utilized the YCSB 95read5update workload with dataset sizes twice and ten times larger than the available memory. The YCSB workload was configured to access documents randomly using a Zipfian distribution. Our findings indicated that decreasing the leaf_page_max setting significantly improved performance, especially on larger instances. For more details, please refer to the JIRA comment.

First, we would appreciate your feedback on whether you have conducted similar tests and observed similar results, and if you agree with the analysis done in PERF-6051.

If your observations align with ours, we recommend considering the use case described in PERF-6051 as part of your leaf_page_max tuning guide.

Definition of done:

• WiredTiger documentation is updated to explain various leaf_page_max values.
• We have decided whether to update the MongoDB documentation and create a DOC ticket.