| Bug #120597 | performance_schema atomic values maintaining CPU cache-friendliness | ||
|---|---|---|---|
| Submitted: | 2 Jun 6:58 | ||
| Reporter: | Alex Zimnitski | Email Updates: | |
| Status: | Open | Impact on me: | |
| Category: | MySQL Server: Performance Schema | Severity: | S5 (Performance) |
| Version: | 8.0.46, 8.4.9 | OS: | Linux |
| Assigned to: | CPU Architecture: | x86 | |
Portions of this website are copyright © 2001, 2002 The PHP Group
Page generated in 0.046 sec. using MySQL 8.4.5-u11-cloud
Timestamp references displayed by the system are UTC.
Content reproduced on this site is the property of the
respective copyright holders.
It is not reviewed in advance
by Oracle and does not necessarily represent the opinion of
Oracle or any other party.
© 2026, Oracle Corporation and/or its affiliates
Description: False sharing is a well-known problem in multiprocessor systems, causing performance degradation in multi-threaded programs running in such environments CPU Cache has a significant impact on performance. Maintaining cache-friendliness /** An atomic @c uint32 variable, guaranteed to be alone in a CPU cache line. This is for performance, for variables accessed very frequently. */ struct PFS_cacheline_atomic_uint32 { std::atomic<uint32> m_u32; char m_full_cache_line[PFS_CACHE_LINE_SIZE - sizeof(std::atomic<uint32>)]; PFS_cacheline_atomic_uint32() : m_u32(0) {} }; struct PFS_cacheline_atomic_uint64 { struct PFS_cacheline_atomic_size_t { ... How to repeat: Run on server with 2 sockets, each socket 64 logical CPU (32 physical), 4 NUMA-nodes. Disabling performance_schema significantly improves performance. Suggested fix: Remove align to full cache line. Separete atomic value to CPU count Aggregate separated values in performance_schema