[MDEV-20416] benchmark the effect of default_character_set=utf8mb4 Created: 2019-08-25  Updated: 2021-11-02  Resolved: 2021-11-02

Status: Closed
Project: MariaDB Server
Component/s: Character Sets
Fix Version/s: N/A

Type: Task Priority: Major
Reporter: Sergei Golubchik Assignee: Axel Schwenke
Resolution: Fixed Votes: 1
Labels: None

Attachments: PDF File MariaDB-10.6.pdf     Text File result.txt    
Issue Links:
Blocks
blocks MDEV-17662 Default to UTF8 Closed

 Description   

run the standard regression-detection-between-releases set of benchmarks for a 10.4 build with the default character set changed to utf8mb4

 Comments   
Comment by Axel Schwenke [ 2019-11-26 ]

I added result.txt containing numbers from the regression tests. The difference between 10.4.8 and 10.4.8-mb4 ist this extra line in my.cnf:

character_set_server=utf8mb4

Observations: no impact on read-only test cases. Read-write tests become slower by ~10..15%. Extreme case is read-write using MEMORY tables with hash-indexes. Here the slowdown is 75%.

One test - 't_writes-binlog-multi' - actually becomes a littlke faster with utf8mb4

PS: you can ignore the single-threaded ARIA-insert tests. Those deliver unstable results all the time.

Comment by Axel Schwenke [ 2021-11-02 ]

MariaDB-10.6.pdf contains a detailed comparison of the different Unicode collations for the implementation in MariaDB-10.6.3. Workload is OLTP range selects (one of each flavor: simple, odered, distinct, sum). utf8mb4 is in most cases even faster than utf8mb3.

Generated at Thu Feb 08 08:59:19 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.