CFR: A cooperative link failure recovery scheme in software-defi...

plane

the switches in the data plane, where the interface connecting the control plane with data plane is called southbound interface.

Among the existing southbound interfaces, OpenFlow

sponsored by Open Networking Foundation (ONF)

is most widely

adopted. In the OpenFlow network, the network traffic is routed by following the flow entries installed into the switches. When

a new packet is transmitted to the switch, it will be forwarded to its destination if the packet head can match any flow rule;

otherwise, a request will be sent to the controller which is responsible for the transmission path.

works. As long as link failures are detected in SDNs, the controller will recover the failed links by installing backup rules into

the switches. One challenge of recovering such link failures is how to perform the fault recovery timely. Another challenge is

how to carefully allocate network resources during the recovery process, because various network applications have different

QoS requirements such as delay and bandwidth.

the recovery. Current recovery schemes including proactive-based and reactive-based recovery ones may be difficult to satisfy

the increasing demands during the recovery due to ignoring the trade-off between resource utilization and recovery time. On

the one hand, the proactive recovery scheme needs additional Ternary Content Addressable Memory (TCAM)

resources for

achieving local recovery. Since an SDN switch can only support limited rules in its TCAM

, the storage resources reserved

for backup paths will be wasted if there is not any failure. Furthermore, the computed backup paths before failure without

considering real-time network status may be inappropriate for the network applications with QoS requirements. On the other

hand, the reactive recovery scheme calculates the backup paths after detecting the link failures, which may cause serious packet

• We are the first that proposes a cooperative link failure recovery scheme to achieve a fine trade-off between resource

• We use a multi-objective optimization model to formalize the link recovery problem in SDNs. To address the problem, we

• We apply fast failover and VLAN features supported in OpenFlow protocol to enhance the performance of CFR, which

• We implement extensive simulations based on Mininet, and evaluate the performance of CFR by comparing with two

and TI-LFA

According to whether the controller intervenes in recovery, existing strategies to recover the link failure can be classified into

two categories: reactive strategy and proactive strategy

for normal operations. In the link failure recovery, reactive strategy installs backup flow entries into the corresponding flow

tables which locate in the switch after detecting the failure. Sharma et al.

reactive strategies, and proposed a reactive recovery scheme to compute the backup paths based on NOX, a OpenFlow supported

controller. The proposed scheme emanated all existing paths to detect the failed link, which reduced the computation efficiency

and increased the burden of the controller. Kim et al.

proposed a failure recovery framework which could recover the failure

and Dijkstra algorithms to restore the network from a link failure. The presented model provided a shortest path with reducing

delay and offered more efficient bandwidth through redundant links. Although reactive restoration strategy has a better failure

recovery with dynamic path computing, it is hard to meet the carrier-grade recovery within 50 ms

requirement in large and

complex networks. In real-time complex networks, Li et al.

safe point. Zhang et al.

the aim to conserve the TCAM space.

detect the failure and detour the disrupted flows to the backup paths locally. Van et al.

exploited the group table feature to

accomplish rapid failure recovery. The proposed scheme was inspired of crankback signaling, a backtracking procedure defined

in RFC, and it could avoid the problem where the neighbor of failed node had no backup routing paths. Although this kind

of method reduced packet loss, it might generate a forwarding loop. Padma et al.

proposed a link protection scheme, which

achieved a local link failure recovery within 50 ms. Although the proposed scheme degraded the burden of controller, it occupied

large resources due to generating large backup rules. Considering the limited TCAM space in OpenFlow supported switches,

Mohan et al.

proposed a TCAM-aware recovery scheme to save the TCAM space by reusing the primary paths and aggregating