Cementing High Availability in OpenFlow with RuleBricks

In this study, the authors proposed a so-called RuleBricks data structure residing in controller and representing (wildcard) forwarding rules to address the limitations caused by the churn of replicas running "behind" Openflow switches (controller replicas). Via modelling rules by RuleBricks, they also leveraged Chord protocol to define how to apply such data structures to elements in SDN network.

Problems

How can HA policies be added to Openflow's forwarding rules?

• When a replica fails, the incoming flow should be reassigned to a designated backup (of course without corrupting session state).
• The fragment of forwarding rules as replicas are created or destroyed

Solution - RuleBricks

Each brick corresponds to a segment of the address space. All flows with source IP in the address range covered by a brick will be forwarded to the same replica.

Operate via three primitives: 

1. Drop: add new active rules
2. Insert: add new backup rules => planning for failure
3. Reduce: make rules more efficient

As in Chord ring, each replica is responsible for objects (flows) that map to a portion of the address space. When a new replica is introduced, it usually requires to drop new bricks to reassign flows to the corresponding replicas.