Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.2.2
-
None
Description
Starting with MariaDB 10.2.2 there is a heavy performance regression for write-only workload and low thread counts.
thread count | 10.2.1 QPS | 10.2.2 QPS | Change |
---|---|---|---|
1 | 7233.9 | 4622.5 | -36% |
8 | 43185 | 32226 | -25% |
16 | 75237 | 61746 | -18% |
32 | 130255 | 114321 | -12% |
64 | 168282 | 173745 | +3% |
128 | 179679 | 205590 | +14% |
256 | 179779 | 209018 | +16% |
Benchmark: sysbench OLTP with --oltp_simple_ranges=0 --oltp-distinct-ranges=0 --oltp-sum-ranges=0 --oltp-order-ranges=0 --oltp-point-selects=0 --oltp_non_index_updates=9 --oltp_index_updates=9. 32 tables with 10 mio rows total (yields ~2.5GB tablespaces). Datadir residing on RAID-0 on two SSD.
my.cnf
[mysqld]
|
performance_schema = 0
|
secure-file-priv = /tmp
|
|
max_connections = 600
|
table_open_cache = 1200
|
query_cache_type = 0
|
|
innodb-file-per-table = true
|
innodb-flush-method = O_DIRECT
|
innodb-buffer-pool-size = 16G
|
innodb_log_file_size = 2G
|
innodb_log_buffer_size = 8M
|
innodb_buffer_pool_instances = 8
|
loose-innodb_adaptive_hash_index_parts = 16
|
loose-innodb_adaptive_hash_index_partitions = 16
|
innodb_io_capacity = 5000
|
loose-innodb_flush_neighbors = 0
|
innodb_write_io_threads = 8
|
This issue popped up during work on MDEV-10123 (see there for more numbers). This could be the same issue as MDEV-11937.
Attachments
Issue Links
- relates to
-
MDEV-21423 lock-free trx_sys get performance regression cause by lf_find and ut_delay
- Stalled
-
MDEV-11937 InnoDB flushes redo log too often
- Closed