-
Type: Bug
-
Resolution: Unresolved
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
Storage Engines
-
2024-05-28 - FOLLOW ON SPRINT
The medium-btree workload shows a visible regression (~5-6%) with the new tcmalloc.
The performance drop at commit 24,461 corresponds to WT-12068, when we enabled the new tcmalloc in WT's performance tests.
The test here creates a 5GB tree and then has 16 threads perform random reads for 2 minutes. Since we've seen similar regressions correlated with the new tcmalloc in the server, it would be useful to know if the same issue is happening here (larger allocations from the kernel leading to a larger memory resident size and less memory in the kernel page cache, leading to more page cache misses). Even if it is a different issue it would be useful to understand it.