← Back to team overview

yahoo-eng-team team mailing list archive

[Bug 1502297] [NEW] Improve SG performance as VMs/containers scale on compute node

 

Public bug reported:

Please refer to the comments in the following bug:

https://bugs.launchpad.net/neutron/+bug/1492456

In which it was suggested to handle improving SG programming performance
as a RFE bug.

To Summarize the problem, when there are about 160 VMs, the neutron-ovs-
agent takes more than 2 seconds per VM  to program the iptables rules
mainly because of the inefficiency in the iptables programming code.

#VMs = 0, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 0.34 After: 0.20

#VMs = 10, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 1.68 After: 0.94

#VMs = 20, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 4.27 After: 2.12

#VMs = 40, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 11.8 After: 6.44

#VMs = 80, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 20.2 After: 13.6

#VMs = 160, provision 10 new VMs on compute node
cumulative time in _modify_rules : Before 50 After: 23.2   <<<<< more than 2 seconds per VM !!!

** Affects: neutron
     Importance: Undecided
         Status: New


** Tags: rfe

** Description changed:

- 
  Please refer to the comments in the following bug:
  
  https://bugs.launchpad.net/neutron/+bug/1492456
  
  In which it was suggested to handle improving SG programming performance
  as a RFE bug.
  
- To Summarize the problem, when there are about 180 VMs/containers, the
- neutron-ovs-agent takes more than 2 seconds per VM/container  to program
- the iptables rules mainly because of the inefficiency in the iptables
- programming code.
+ To Summarize the problem, when there are about 160 VMs, the neutron-ovs-
+ agent takes more than 2 seconds per VM  to program the iptables rules
+ mainly because of the inefficiency in the iptables programming code.
  
  #VMs = 0, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 0.34 After: 0.20
  
  #VMs = 10, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 1.68 After: 0.94
  
  #VMs = 20, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 4.27 After: 2.12
  
  #VMs = 40, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 11.8 After: 6.44
  
  #VMs = 80, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 20.2 After: 13.6
  
  #VMs = 160, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 50 After: 23.2   <<<<< more than 2 seconds per VM !!!

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1502297

Title:
  Improve SG performance as VMs/containers scale on compute node

Status in neutron:
  New

Bug description:
  Please refer to the comments in the following bug:

  https://bugs.launchpad.net/neutron/+bug/1492456

  In which it was suggested to handle improving SG programming
  performance as a RFE bug.

  To Summarize the problem, when there are about 160 VMs, the neutron-
  ovs-agent takes more than 2 seconds per VM  to program the iptables
  rules mainly because of the inefficiency in the iptables programming
  code.

  #VMs = 0, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 0.34 After: 0.20

  #VMs = 10, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 1.68 After: 0.94

  #VMs = 20, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 4.27 After: 2.12

  #VMs = 40, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 11.8 After: 6.44

  #VMs = 80, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 20.2 After: 13.6

  #VMs = 160, provision 10 new VMs on compute node
  cumulative time in _modify_rules : Before 50 After: 23.2   <<<<< more than 2 seconds per VM !!!

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1502297/+subscriptions


Follow ups