[MDEV-25124] benchmark 10.6 performance for PMEM enabled builds Created: 2021-03-12  Updated: 2023-02-27  Resolved: 2021-03-12

Status: Closed
Project: MariaDB Server
Component/s: Storage Engine - InnoDB
Fix Version/s: N/A

Type: Task Priority: Major
Reporter: Axel Schwenke Assignee: Axel Schwenke
Resolution: Fixed Votes: 0
Labels: None

Attachments: PDF File sysbench.pdf    
Issue Links:
Relates
relates to MDEV-14425 Change the InnoDB redo log format to ... Closed
relates to MDEV-17084 Optimize append only files for NVDIMM Closed
relates to MDEV-25090 Install libpmem-dev in the CI environ... Closed

 Description   

Before enabling PMEM in builds by default (cmake -DWITH_PMEM=ON), we want to make sure it doesn't have negative impact on performance on non-PMEM systems.

 Comments   
Comment by Axel Schwenke [ 2021-03-12 ]

Benchmark has been run. Workload: all types of sysbench OLTP workloads, with and without explicit transaction boundaries. Datadir was located on /dev/shm and innodb_flush_log_at_trx_commit = 1 to make sure we test the effects of code changes and not IO limitations of the environment.

Comment by Marko Mäkelä [ 2023-02-27 ]

The sysbench.pdf matches what I observed when running performance tests after implementing a PMEM friendly redo log format and interface in MDEV-14425. With MDEV-14425, I observed some 10% to 15% improvement. I compared the speed of ib_logfile0 stored in the same PMEM device with and without mount -o dax.

Generated at Thu Feb 08 09:35:20 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.