High Availability support

[Romain_JACOTIN]

What kind of high availability is supported by the SDN controller ?

• 1 + 1 ? ( active/passive )
• 1 + 1 ? ( active/active with load-sharing and statefull synchronization capabilities )
• N+ M ? ( all actives with load-sharing and statefull synchronization capabilities )

What are network constraints for high availibility ?

• same L2 subnet for all the controller ?
• Controller can be in different L3 ?

If load-sharing capability is possible, do the SDN controller offer natively distributed statefull capabilities with the use of the JAVA native API ( for SDN app create by tiers ) or is it the responsibility of tiers developpers to handle statefull synchronization between controllers ( with PAXOS stuff for example ) ?

Best regards,

Romain

[Dave_Hawley]

Hi Romain,

The controller supports a 2n+1 database resiliency model.  All controllers are active, and support the ability to assign each OF switch to a specific controller.  The controllers synchronize among each other using L3.

Applications that use the REST API or do not leverage the database infrastructure run on the master controller.  The internal JAVA API supports access to the database for state synchronization, allowing applications to take advantage of the active-active architecture.

Regards, Dave

(HP)