CENTraLISING TraFFIC - DIRK VAN DEN BORNE
Carrier SDN: Centralised traffic engineering in multi-layer networks
Coordination between a network’s transport and IP/ MPLS layers is a promising way to optimise packet-optical networks. But effective integration of the two layers is necessary to avoid isolating them like ships that pass in the night, writes juniper’s Dirk van den Borne.
mUlti-lAyeR cooRDinAtion USe cASeS The IP/MPLS layer leverages the abstract topology to improve both network visualisation and path computation. Multi-layer topology visualisation allows for a consistent visualisation of both network layers, which can be very beneficial for root-cause analysis in case of network failures. Using timestamp information any changes to the transport topology can be communicated up-front to the IP/MPLS layer, allowing for hitless re-calculation and re-signaling of paths. This can facilitate multi-layer topology optimisation use cases, such as bandwidth calendaring or maintenance events. The automated exchange of SRLG information enables the IP/MPLS layer to compute end-to-end disjoint paths taking into account any resources sharing on the transport layer. When the SRLG information changes on the transport layer, for example due to a restoration event, the abstract topology is automatically updated. In addition, the transport controller can communicate to the IP/MPLS controller if a particular abstract link is already protected on the transport layer, or not, which allows the IP/MPLS controller to optimise the use of protection capacity across network layers.
limit the information exchange to only the minimum set of information that is directly useful to improve path calculation. This can be achieved by summarising the transport layer topology into abstract links, which define the end-to-end reachability across the transport layer as well as SRLG information. Any more detailed information, such as network element connectivity or optical transmission impairments, does not improve traffic engineering accuracy and therefore does not need to be exchanged between transport and IP/MPLS layer. To facilitate standardisation the abstract topology is described in a standardised YANG model, which is co-authored by major transport equipment vendors. The abstract topology is then exchanged via either a RESTCONF or REST interface between both network controllers. This allows for an open standards- based approach that relies on programmatic interfaces and lightweight data models that are straightforward to implement.
optimisation. This relies on the exchange of abstract topology information, thereby providing a highly scalable approach for multi-layer coordination. netWoRK toPology ABStRAction A key requirement for multi- layer service provisioning is the capability to install end-to-end diverse paths in the network. Since a centralised controller knows the full IP/MPLS network topology it is capable to compute diverse paths across the network. However, to guarantee that paths are truly diverse it also needs to ensure diversity on the transport layer. When a pair of IP/MPLS circuits (partly) traverses the same nodes or links on the transport layer they will be fate sharing, which is typically referred to as a Shared Risk Link Group (SRLG). This lack of visibility across network layers tends to result in more complicated service provisioning and disconnected planning processes, which indicates there is a benefit to exchange topology information. At the same time, to reduce complexity it is desirable to
DIrK vaN DEN BorNE Senior Architect at Juniper networks M ulti-layer coordination between the transport and IP/ MPLS layers of a network is one of the most promising approaches to build more optimised and simpler multi- layer packet-optical networks. However, both network layers are typically built using strongly different technologies, each with their own approach and legacy for control and management. The recent development of SDN controllers that control either the transport or IP/ MPLS network of the network enables new opportunities for multi-vendor multi-layer packet-optical coordination. The Juniper NorthStar SDN controller is a stateful path computation element (PCE), while the network elements act as path computation clients (PCC). The standardised Path Computation Element Protocol (PCEP) enables communications between the controller and clients. As an active stateful PCE, NorthStar is synchronised with the network in real-time, and therefore has knowledge of both network topology as well as already installed paths and bandwidth reservations. This enables centralised control and optimisation of the network and therefore fits well to software defined network principles. Juniper’s NorthStar SDN controller interfaces directly with one or more transport SDN controller for multi-layer network
Q
V
Q
V
R
R
SRLG
SRLG
W
W
NorthStar Controller
Transport Controller
Yang Abstract Network Topology (REST / RESTCONF via Northbound interface)
SNMP, OpenFlow, NetConf, etc... (Southbound interface to NE)
PCEP, BGP-LS (Southbound interface to NE)
Used Abstr. link
Abstr. link
Packet LSP
Optical path
northStar multi-layer coordination
30
ISSUE 6 | Q1 2016
Made with FlippingBook - professional solution for displaying marketing and sales documents online