Network Orchestration for Dynamic Network Slicing for Fixed and Mobile Vertical Services 1
Rodolfo Alvizu, 2Sebastian Troia, 1Van Minh Nguyen, 1Guido Maier, 1Achille Pattavina (1) SWAN networks, Via Fabio Filzi 27, Milan 20124, Italy (2) Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, Milan 20133, Italy {rodolfo.alvizu, guido.maier, achille.pattavina}@swan-networks.it,
[email protected]
Abstract: We demonstrate how a hybrid and hierarchical transport-SDN control plane based on a network orchestrator and an SDN controller can provide dynamic network slicing for enterprisenetworking services and mobile metro-core networks. OCIS codes: (060.4250) Networks; (060.4256) Networks, network optimization
1. Overview Ten years ago, Software Defined Networking (SDN) started a revolution across the whole networking field with the promise to make network infrastructure as dynamic as other Information Technology (IT) areas, and to break the vendor lock-in over equipment. Today, SDN is mostly deployed at the edge of transport networks, such as in datacenters and campus networks. The transport or core segment is a more challenging context for the penetration of SDN: a greater heterogeneity of equipment specialized for long-haul transmission, and the presence of complex legacy control-plane solutions (e.g. GMPLS) are the major barriers. Despite that, migration to SDN seems the most promising solution for telecommunication service providers (CSPs), as their current transport networks cannot cope with the agility and innovation pace required to meet near future services and applications. Extending the work presented in [1], we demonstrate a hierarchical Transport-SDN (T-SDN) architecture that dynamically creates, modifies and prioritizes network slices over fixed (up to the customer premises entities: CPEs) and mobile (up to the central offices: COs) structurally-converged transport infrastructure. As depicted in Fig. 1, the network orchestrator coordinates provisioning of services. Services with very different constraints on bandwidth, delay and availability may be provided by setting up service-specific network slices. The allocation of resources to the slices dynamically adapts to the requirements and the state of the network. Slicing is a fundamental component of 5G architecture, it and can be offered by CSPs to small, medium and large companies to enable new dynamic enterprise services, and will be the key to dramatically reduce the cost-per-bit of connectivity services. We refer to a hierarchical T-SDN control plane, composed by SDN controllers and a network orchestrator on top (see Fig. 1): this is the best approach to handle the complexity and heterogeneity of transport networks [2]. The hierarchical T-SDN allows to deploy SDN technology in the transport network by integrating it with protocols coming from the legacy control plane, reducing the replacement costs. As shown in Fig. 1, the SDN controllers interface only with CPEs, COs and Provider Edge (PE) nodes, while the control plane of Provider (P) nodes (or core nodes) remains distributed and based on distributed control protocols, such as RSVP, ISIS and BGP. The southbound interface we adopted are BGP-LS, PCEP and NETCONF, while OpenFlow is not needed.
Figure 1: Network Orchestration architecture for dynamic network slicing
2. Innovation We demonstrate a hierarchical T-SDN control plane that dynamically creates, modifies and prioritizes end-to-end connectivity, based on guaranteed MPLS tunnels, that provide customizable virtual topologies. The orchestrator
configures the virtual topologies, access lists and policy based routing to provide dynamic network slices. The customizable slices provide different degrees of service level agreement (SLA) in terms of bandwidth, delay and availability requirements, and are associated through access lists (based on protocol, port, destination) to specific types of traffic or services at the CPEs and COs. The SDN control plane introduces two degrees of flexibility of the tunnels: path and bandwidth adaptation, so the resource allocation adapts to network conditions and bandwidth requirements. The amount of guaranteed bandwidth requested by the end-users (or by the service) can be modified on demand via web interface (or via API). This is achieved by defining a bandwidth calendar that specifies the required bandwidth at each hour of the day. The demo is presented in a virtualized environment provided by Cisco VIRL. The network is composed by two flavors of router images: IOSv (CPE, CO and P nodes) and IOS XRv (PE nodes). NETCONF/CLI and NETCONF/YANG are used to configure the network elements, BGP-LS to obtain link-state information and Path Computation Element Protocol (PCEP) to install MPLS tunnels. Two modes will be available to set up the tunnels: a) using traditional MPLS signaling; b) using Segment Routing (SR): the latter solution ensures more scalability when the number of slices to manage increases. OpenDaylight (ODL) SDN controller, provides several southbound interfaces (SBIs) to control and configure network elements such as: NETCONF/YANG (RFC 6241), BGP-LS (RFC 7752) and PCEP (RFC 5440). ODL requires to mount a YANG data model of the devices to be controlled. However, the IOSv does not support YANG models: as consequence, ODL’s NETCONF SBI cannot be used to configure IOSv nodes. Therefore, the orchestrator implements a NETCONF/CLI interface to configure IOSv nodes (see Fig. 1). The network orchestrator provides REST APIs for applications, such as the Telco-Dashboard (part of the demo) and the end-users’ web portal to configure and control their network services. The GUI allows a service provider to explore the temporal evolution of the dynamic resource allocation in the network. This is an interesting tool to assess the impact of traffic changes over the resource allocation in the network, and makes managing simple and effective. The demonstration will be done live on the virtualized lab created with Cisco VIRL, in the framework of a collaboration project between Politecnico di Milano and SWAN networks. We designed the following use-cases to demonstrate the benefits of dynamic creation and modification of network slices: i) Enterprise network use-case: composed by two companies with several locations and diverse per-service network slices. Traffic is injected following per-slice bandwidth calendars. ii) Mobile metro-core network slice use-case: composed by the radio access network connected to CO nodes and mobile metro-core. Traffic is injected at COs following the patterns from specific land usage regions in a metropolitan area (see Fig. 1) that were extracted in [3]. iii) A video streaming use-case: demonstrates the effects of dynamic network slices on delay sensitive services when reconfiguration occurs to adapt to changes on network state and/or requirements. Machine learning can be used to improve future decisions on resource allocation as in [4]. Accurate predictions of traffic can further improve such decisions. In this demo, traffic patterns are introduced via the bandwidth calendar available at the web interface as in [1], and extracted from network history as in [3]. In the virtualized lab environment, of this demo, we assume that the MPLS layer is directly mapped onto the optical layer, as according to the virtual wavelength path model (VWP) [4]. In VWP each node has grooming capability, and packet traffic is groomed into wavelengths at every hop. 2. OFC Relevance We demonstrate the use of orchestration to provide network slices that can fulfill the requirements of future fixedmobile converged networks that must support 5G and fixed vertical services. The use of dynamic tunnels allows to optimize the network resources, that can serve more customers and services without over provisioning. Even by assuming a VWP model, our approach allows to reduce the power consumption in the optical layer, by allowing to reduce the number of active wavelengths (in consequence optical amplifiers), and to reduce the number of transponders needed to serve the requests. 4. References [1] R. Alvizu, G. Maier, S. Troia, Van Minh Nguyen and A. Pattavina, "SDN-based network orchestration for new dynamic Enterprise Networking services," in ICTON, Girona, 2017, pp. 1-4. [2] R. Alvizu et al., "Comprehensive survey on T-SDN: Software-defined Networking for Transport Networks," in IEEE Communications Surveys & Tutorials, vol. PP, no. 99, pp. 1-1. [3] S. Troia et al, "Identification of tidal-traffic patterns in metro-area mobile networks via Matrix Factorization based model," 2017 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), Kona, HI, 2017, pp. 297-301. [4] R. Alvizu, S. Troia, G. Maier and A. Pattavina, "Matheuristic with machine-learning-based prediction for software-defined mobile metro-core networks," in IEEE/OSA Journal of Optical Communications and Networking, vol. 9, no. 9, pp. D19-D30, Sept. 2017.