ACE4710 - persistence rebalance performance

[Surya ARBY]

Hello all.

When enabling persistence rebalance, does it force the traffic to go to the control plane or does it stay on the Cavium Octeon chipset ?

[amacuz · Accepted solution]

Hi Surya,

the connection is not completely proxied all the time and of course is not totally un-proxied

I think you can find a good explanation of the mechanism in this document:

http://www.cisco.com/en/US/prod/collateral/modules/ps2706/ps6906/White_Paper_Connection_Handling_within_the_Cisco_Application_Control_Engine_Module_Hardware_ps7027_Products_White_Paper.html

Search for the sentence "This process is called reproxy" within the document and you'll find the paragraph that should be of interest for you.

I hope this clarifies,

Alessandro

--
If  this helps you and/or answers your question please mark the  question as  "answered" and/or rate it, so other users can easily find  it. Thanks

[Surya ARBY]

Thank you.

So as I understand, no load balanced traffic hits the control plane (except the flows on which you perform advanced HTTP optimizations like Smart redirect, Flash forward...) I've always believed that enabling eprsistence rebalance would kill the CPU by sending all he traffic of the class-map to the CP.

[amacuz]

You are welcome  Surya,

ideally after the decision is taken (on which rserver to select and this happens always in the first packets of the flow), the connections should stay as much as possible in the fast-path.

Unless, based on the conditions and the environment and the application loadbalanced, you configure the ACE differently.

Alessandro