29-08-2016, 11:52 AM
1451376656-IndependentTransientPlaneDesign.pdf (Size: 1.38 MB / Downloads: 4)
Abstract—Network protection against link failure is required
to allow packets to reach the destination with minimal
packet loss when a failure occurs. When a link fails,
traffic that attempts to use the failed link is interrupted.
Typically, routers in the network discover the failure and
find a new route to bypass the failed link. Alternatively,
well-known segment protection schemes can also be used
to speed up the link recovery time by rerouting packets locally
through precalculated protection paths. However,
several backup paths have to be prepared for each primary
path, making path configuration rather complex and
poorly scalable. This paper proposes a design for fast rerouting
in an OpenFlow-based network. This new design
reduces the number of flow entries and the number of configuration
messages needed for network rerouting, which
in turn reduces the memory size needed in each switch
and the CPU load at the controller. We show empirically
and using simulations that our design can reduce the number
of flow entries and configuration messages needed by
about 60% and 75%, respectively, when compared with an
existing OpenFlow-based segment protection design. Furthermore,
we implement the proposed design on a panEuropean
network and show that our design can recover
from a link failure in as little as 25 ms.
Index Terms—Mininet; Network protection; OpenFlow;
Routing; Software-defined network.
I. INTRODUCTION
F ailure protection is an important issue in network survivability.
When connectivity is lost due to a failure,
packets that attempt to pass the failed link cannot be delivered
to their destination. An open shortest path first
(OSPF) network, in which a link-state-based routing protocol
is used to learn network topology, can detect the link
failure and converge to a new topology. However, recovery
time in the OSPF network takes more than a second [1].
Recovery times of this order are not acceptable in many
networks, as target protection times of 50 ms or 100 ms
are common, respectively, for leased line traffic and
video/audio services [2]. Multiprotocol label switched
(MPLS) networks provide fast rerouting using an alternative
label switched path (LSP) to reroute packets from a node connected to the failed link to another node or to
the destination. In a typical MPLS network, operating a
distributed control plane [3,4], forwarding decisions are
made locally by each switch, based on its predefined configurations.
Although this allows for dynamic and automatic
forwarding decisions to be made in the local
switch, there is no guarantee that the chosen route is still
optimal when the failure occurs.
Segment protection approaches rely on preplanned
backup paths that deflect packets from a primary path to
other paths [5–7], as shown in Fig. 1. At each switch, primary
and backup ports are defined. Packets travel through
the primary path via the primary port of each switch to the
destination during normal operations. If the switch detects
that the primary port is not available, the packets are deflected
from the primary path to a neighboring switch via
the backup port of that switch. The neighboring switch then
forwards the packets to the destination via its primary port.
Software-defined networking (SDN) [8,9] is an attractive
solution that enables a network administrator to control
the network requirements, including how to deal with failure
protection [10,11]. SDN separates the network control
and data forwarding planes. This allows network engineers
and administrators to respond quickly to changing requirements
of the network from a centralized controller. The
network administrator can avail of improved network programmability
to flexibly control the physical switches from
the controller, which runs all the intelligent control and
management software, regardless of the vendor or the
model of the switches used.
OpenFlow [12,13] is a widely known protocol (southbound
interface) for SDN networks that enables the controller
to interact with the forwarding plane of the switches and
thus make adjustments to the network. The hardware
switches can also use OpenFlow messages to inform the controller
when links go down or when a packet arrives with no
specified forwarding instructions. The forwarding instructions
are based on a flow entry, which is defined by a set
of specific parameters, such as the source and destination
Ethernet/IP addresses, the switch input port and VLAN
tag, etc. [13]. The controller specifies the set of parameters
and how packets that match the flow entry should be processed.
Packets are matched against flow entries based on
prioritization. Higher priority flow entries are used, when
http://dx.doi10.1364/JOCN.7.000264 available, over lower priority ones.