[MXS-731] Optimise the use of spinlock while draining the write queue Created: 2016-05-18  Updated: 2017-12-01  Resolved: 2016-05-31

Status: Closed
Project: MariaDB MaxScale
Component/s: Core
Affects Version/s: 1.4.3
Fix Version/s: 2.0.0

Type: New Feature Priority: Major
Reporter: martin brampton (Inactive) Assignee: martin brampton (Inactive)
Resolution: Done Votes: 0
Labels: None

Issue Links:
Relates
relates to MXS-721 Improve scalability at multi threaded... Closed

 Description   

Access to the write queue, which is a linked list of buffers, results in contention. This can cause spinlocks to spin for excessive periods. Modifications are needed to make the locked sections of code much shorter .



 Comments   
Comment by martin brampton (Inactive) [ 2016-05-31 ]

This code demonstrates significant gains in profiling tests, and is now merged into the develop branch, for release as part of version 2.0. The gains will vary considerably according to different patterns of usage.

Generated at Thu Feb 08 04:01:30 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.