[OPNFLWPLUG-254] cbench data store performance very poor Created: 26/Aug/14  Updated: 27/Sep/21  Resolved: 18/Nov/14

Status: Resolved
Project: OpenFlowPlugin
Component/s: General
Affects Version/s: None
Fix Version/s: None

Type: Bug
Reporter: Jan Medved Assignee: Jan Medved
Resolution: Done Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified
Environment:

Operating System: Mac OS
Platform: Macintosh


External issue ID: 1626

 Description   

The performance of the cbench loopback through the data store (dropAllPackets) is very bad (1-2 operations per second). The poor performance is not related to the data store itself, as other data store benchmarks keep performing well.

To reproduce, issue 'dropAllPackets on' on the osgi console and run the cbench test.



 Comments   
Comment by Michal Rehak [ 03/Sep/14 ]

The performance is not as poor as it looks: there are for example 400 flowMod messages reported by cbench but we sent 50000 of them. Problem is that cbench does not count floMod with modify command but only with add command.

I suspect that new dataStore API introduced this behavior:

Write 1. flow into DS/config:

  • get corresponding addFlow dataChangeEvent
    Write 2. flow into DS/config:
  • get corresponding addFlow dataChangeEvent
  • get update for 1. flow (original == updated flow)
    Write 3. flow into DS/config:
  • get corresponding addFlow dataChangeEvent
  • get update for 1. flow (original == updated flow)
  • get update for 2. flow (original == updated flow)
    etc..

So the amount of flowMod messages physically sent to node grows linear and for 320 flows we should have 51360 messages sent.

This can be simulated in both openflowplugin distributions (cbench, base) regardless the dropAllPacket* is turned on or off. And with mininet running on OF-1.0 or OF-1.3.

Comment by Abhijit Kumbhare [ 17/Sep/14 ]

Michal to retest.

Comment by Jan Medved [ 18/Sep/14 ]

Well, this would mean that we create a flood of flowmods for every new flow added - that's still a very bad behavior

Comment by Michal Rehak [ 22/Sep/14 ]

https://git.opendaylight.org/gerrit/#/c/11458/

Comment by Michal Rehak [ 18/Nov/14 ]

no response

Generated at Wed Feb 07 20:31:58 UTC 2024 using Jira 8.20.10#820010-sha1:ace47f9899e9ee25d7157d59aa17ab06aee30d3d.