Software Defined Networking (SDN)

• Several challenges became pronounced as IP networks became more adopted:
• Handling complexity and the dynamic nature of networks
• Tightly coupled architecture, making flexibility difficult
• Software Defined Networking (SDN) attempts to overcome the limitations of the legacy IP networking paradigm.
• Separation of control logic from the data plane is key.
• Network switches simply forward, and control logic is in a logically centralized controller.
• Allows innovation in network reconfiguration and policy enforcement.
• Production-level SDNs need a physically distributed control plane for performance, reliability, and scalability.
• Achieved by using a programming interface between the SDN controller and the switches.
• SDN principles allow separation of concerns between networking policies, their hardware implementation, and traffic forwarding.
• Newer networking abstractions & simplifying network management enable innovation.
• Traditional computer networks have three abstract logical planes of functionality:
• Data plane: Forwards data in packets or frames.
• Control plane: Determines which path to use.
• Management plane: Monitors and configures the control functionality via tools like SNMP.
• Conventional networks have tightly coupled data and control planes.
• New features require modifying all control plane devices like installing new firmware/hardware.
• Specialized equipment called "middleboxes" for functions like load balancers, intrusion detection, and firewalls but are hard to reconfigure.
• Since SDN decouples the control plane, middle box services can be viewed as an SDN controller application with several advantages:
• Shared abstractions.
• Consistency of network information.
• Locality of functionality placement.
• Simpler integration.
• The SDN landscape can be decomposed into layers. Each layer has its own functions through different technologies.
• SDN Infrastructure consists of networking equipment (routers, switches, middlebox hardware).
• Equipment are merely forwarding elements, operated by a centralized control system.
• OpenFlow (software) switches are examples include SwitchLight, Open vSwitch, and Pica8.
• Southbound interfaces act as connecting bridges between connecting and forwarding elements also control and data planes.
• They are coupled with the forwarding elements of the underlying physical or virtual infrastructure.
• Most popular is OpenFlow, with others like ForCES, OVSDB, POF, OpFlex, OpenState, etc.
• For complete network virtualization, network infrastructure needs to support arbitrary network topologies and addressing schemes.
• New advancements in SDN network virtualization such as VXLAN, NVGRE, FlowVisor, FlowN, and NVP.
• Network operating systems (NOS) ease network management and solve problems with a logically centralized controller.

Network Operating Systems (NOS)

• A NOS provides abstractions, essential services, and common APIs.
• Examples of NOSs are OpenDayLight, OpenContrail, Onix, Beacon and HP VAN SDN.
• Northbound interfaces are the other core abstraction to Southbound interfaces.
• Standard for Northbound interface is still an open problem and so are its use cases.
• Requires abstraction that guarantees programming language and controller independence.
• Examples are Floodlight, Trema, NOX, Onix and SFNet.
• Language-Based virtualization expresses modularity and many levels of abstraction.
• Takes complexity away from application developers without compromising security.
• Examples are Pyretic, libNetVirt, AutoSlice, RadioVisor, OpenVirteX, etc.
• Network programmability can be achieved using low-level or high-level programming languages.
• Higher-level languages provide abstractions, are more modular, code, more reusable in control plane.
• Examples are Pyretic, Frenetic, Merlin, Nettle, Procera, and FML.
• Network applications implement control plane logic and translate to commands in the data plane.
• SDN applications are routing, load balancing, security enforcement, end-to-end QoS power consumption reduction.
• Some well-known solutions are Hedera, Aster*x, OSP, OpenQoS, Pronto, Plug-N-Serve, SIMPLE, FAMS, FlowSense, OpenTCP, NetGraph, FortNOX, FlowNAC, and VAVE.
• SDN infrastructure includes networking equipment, physical devices that do not have embedded intelligence or control.
• Networks are built on top of open/standard interfaces that ensure configuration.
• The SDN architecture device forwards packets and a controller is a software stack.
• A model derived from OpenFlow is widely accepted for SDN data plane devices including:
  • A matching rule.
  • Actions to be executed.
  • Counters.
• Other SDN-enabled forwarding include Protocol-Oblivious Forwarding (POF) and Negotiable Datapath Models (NDMs).
• In an OpenFlow device, when a packet arrives, the process starts in the first table until it either matches or misses. Some actions include:
• Forward the packet.
• Encapsulate the packet and forward it to controller
• Drop the packet
• Send the packet to normal processing pipeline
• Send the packet to next flow table
• Southbound interfaces or APIs provide the medium the control and data plane work.
• An API, proposals like OpenFlow have received reception, because they promote interoperability of vendor-agnostic devices.
• Currently, OpenFlow is widely accepted for SDNs and provides specification to implement devices, along with communication channel between data and control plane.
• There are three information sources from OpenFlow protocol.
• Event-based messages sent by forwarding devices to controller when there is a link or port change.
• Flow statistics are generated by forwarding devices and collected by controller.
• Packet messages are sent by forwarding devices to controller when they do not know what to do with a new flow.

Southbound Interfaces and SDN

• Despite OpenFlow, there are API proposals such as ForCES, OVSDB, POF, OpFlex, OpenState, etc.
• ForCES provides flexible approach to traditional network management while not needing a logically centralized controller and the control and data planes can be kept on the same network element.
• OVSDB is a complement to OpenFlow or Open vSwitch and allows control elements to create multiple vSwitch instances and set QoS.
• Controllers solve the challenge of device-agnostic developments and abstraction through means of a logically centralized control.
• SDN Controllers can be categorized based on centralized or distributed architecture.
• Centralized controllers use a single entity to manage all forwarding devices in the network. However these have issues.
• Some enterprise networks and data centers use architectures, such as Maestro, Beacon, and NOX-MT. Additionally, Multi-threaded designs are used to explore how multi-core computer architectures handle the load.
• Single controller architectures such as Trema, and Ryu NOS target environments such as data centers, cloud infras.
• Single controller Rosemary offer specific functionality, guarantees security and application isolation by using a container based architecture called micro-NOS.
• Unlike single controller architectures a distributed network controller can be scaled to meet network requirements of small or large networks.
• Distribution has two forms:
• Centralized cluster of nodes or physically distributed set of elements.
• Cloud providers which run multiple data centers interconnected by a WAN may require a hybrid approach.
• Properties of distributed controllers:
• Weak consistency semantics
• Fault tolerance
• Open Networking System (ONOS) aims to provide a global view, scale-out performance and fault tolerance.
• Was built on Floodlight, an open-source single-instance SDN controller.
• Several ONOS instances run in a cluster, which share network state for a global network view.
• OpenFlow managers receive changes the applications make to the view, and the switches are programmed.
• Titan (graph database) and Cassandra (distributed key value store) is used to implement the view using a Blueprints graph API.
• Each ONOS instance serves as the master OpenFlow controller and in case of failure, ONOS redistributes any other remaining instances.
• Each switch connect to multiple ONOS instances and upon failure an election is held to choose a master for each of the switches.
• Zoopkeeper is used to maintain the mastership between the switch and the controller.
• P4 (Programming Protocol-independent Packet Processors) is a high-level programming language to configure switches in conjunction with SDN control protocols to better expose a switch's functionalities to the controller.
• P4 is used to configure the switch programmatically and acts as a general interface between the switches & controller.
• Primary goals of P4:
• Reconfigurability: The way parsing and processing of packets in the switches should be modifiable by the controller.
• Protocol independence: Enables switches to be independent of protocols & the controller defines a packet parser.
• Target independence: Programs written in P4 should be converted into target-dependent programs by a compiler.

P4 Packet Programming Overview

• The switches using P4 use a programmable parser and a set of match+action tables to forward packets.
• Tables can be accessed in multiple stages contrasting OpenFlow, which supports only fixed parsers.
• The P4 model enables packet processing across various forwarding devices such as routers, load balancers, NPUs, etc.
• Operations of the P4 forwarding model:
• Configure: Program the parser by specifying the header fields to be processed in each match+action stage, and defining the order of these stages.
• Populate: Alter the entries in the match+action tables specified during configuration.
• Configuration determines packet processing and the supported protocols, population decides the policies to be applied.
• P4 is one such packet processing language where:
• Legal header types are declared and allow the parser is aware of the possible packet formats.
• A control flow program uses the declared header types and a set of actions to specify processing of the headers.
• Table Dependency Graphs (TDGs) are used to identify the dependencies between the header fields and help determine the order in which the tables can be executed.
• P4 programming language is used to address the dependency issues and it is fed into a compiler for dependency analysis by translating those results in TDG logic.

Traffic Engineering

• The major applications areas of SDN are overviewed.
• Interest in optimizing traffic flow to reduce power consumption, use network resources and perform load balancing.
• Monitoring plane load via southbound interfaces.
• ElasticTree (identifies and shuts down links and devices).
• Plug-n-Serve and Aster*x (achieves scalability by handling of large numbers of requests).
• Automate management of router configuration to reduce the growth with reducing routing tables so information does not duplicate.
• SDN optimizes traffic at scale using dynamic provisioning of VPNs in cloud infrastructure (ALTO VPN)
• Existing wireless networks face challenges and some wireless networks (WLANS, cellular networks) and management are made easier using SDN.
• Wireless networks give on-demand virtual access points (VAPs), dynamic usage of spectrum and sharing of wireless infrastructure.
• OpenRadio enables decoupling of the wireless protocols from the underlying hardware by providing an abstraction layer.
• Virtual access points (LVAPs) give managing wireless networks by mapping of LVAPs and clients and the Odin framework uses these functions to allow users to seamlessly move between APs.
• Application aims to add features to other networking services. New functions can be added to systems such as BISmark which enables detection/response in network conditions.
• These applications improve the existing features of SDNs using OpenFlow which reduces the load arising from collection of data plane statistics.
• OpenSketch (southbound API). Some examples include:
• OpenSample and PayLess (monitoring frameworks).
• Improves the security of networks.
• Uses devices and enforcement policies on networks and DDoS mitigation from information collected.
• Detection of anomalies and monitor traffic through the cloud.

Data Center Networking and SDX Infrastructure

• SDN is for detecting anomalies and has many applications for live migrations etc.
• Example of "Software Defined Internet Exchange"
• SDN improve the operation of an IXP.
• Used for the routing of packets across the Internet through the Boarder Gateway Protocol (BGP).
• The main limitations of the BGP are:
• Routing on destination IP prefix - The route is the destination of the incoming packet and can not be customized for the traffic.
• Networks have control end-to-end with indirect mechanisms.
• Using SDN Researchers propose addressing BGP.
• Conduct actions by matching over various header fields.
• SDN, physical location. Facilitates exchanges where researchers have proposed an SDN. SDX applications were proposed to implement including:
• Custom Peering
• Traffic inbound is on source forward. Destination rewrites. Target traffic.

SDX infrastructure and Applications

• In SDX, Each AS can define forwarding policies by simply assuming it's the only participant at the SDX but it does not influence others virtual switches.
• Policies can also be different based on the direction of the traffic in and out to the virtual switch and also from the virtual switch data from other participants
• Example A and B's traffic as expressed in match statements
• Wide Application; Traffic Delivery addresses the need to improve application to multiple areas that need delivery improvements:
• ISPS prefer dedicated ASes.
• SDN enables packets to be defined to enable and control the traffic to its destination for custom reasons.
• The existing approaches using DNS is a lower server, but is caching responses are cases of failures can be improved by using SDN that supports the modification of DNS with a 100% success rate. Addressed the challenges and existing approaches use the placements of the middle boxes and is targeted to direct those services to a sequence of middle boxes.