Native OVSDB transation commit shows O(n) performance
Bug #1499893 reported by
Ryan Moats
This bug affects 2 people
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
neutron |
Won't Fix
|
Low
|
Ryan Moats |
Bug Description
Create 100 tenants each one with the following setup where each router is scheduled to the same legacy node that has the L3 agent configured to use the native OVSDB inerface.
tenant network ------- router ------ external network
Reference http://
In the above graph, y-axis is time in seconds, and x-axis is pass through _ovs_add_port (two per router add).
DbSetCommand's performance increases with each router add. To support scale, this needs to be closer to O(1) and perform significantly better than using ovs-vsctl via rootwrap daemon.
Changed in neutron: | |
importance: | High → Medium |
tags: | added: kilo-backport-potential liberty-rc-potential |
tags: |
added: loadimpact removed: performance |
Changed in neutron: | |
assignee: | nobody → Ryan Moats (rmoats) |
tags: |
added: liberty-backport-potential removed: liberty-rc-potential |
tags: | added: mitaka-backport-potential |
tags: | added: scale |
tags: | removed: kilo-backport-potential liberty-backport-potential |
tags: | removed: mitaka-backport-potential |
To post a comment you must log in.
The above graph incorrectly assigns the transaction commit time to DbSetCommand. The corrected graph is here: http:// ibin.co/ 2HEOucFKZBOv