Institute of Computer Science Department of Distributed Systems Prof. Dr.-Ing. P. Tran-Gia SDN Interfaces and Performance Analysis of SDN components, David Hock, Michael Jarschel, Thomas Zinner, Phuoc Tran-Gia www3.informatik.uni-wuerzburg.de
Agenda u A Compass for SDN Interfaces Features Use cases u Performance of the SDN architecture Data Plane Performance Control Plane Performance Analytical Model u Network Functions Virtualization Placement in a Mobile Network Performance Evaluation a virtualized network function 2
IEEE Communications Magazine, June 2014 M. Jarschel, T. Zinner, T. Hossfeld, P. Tran-Gia, W. Kellerer A COMPASS FOR SDN 3
Interfaces Applica8on Control Module Applica8on Control Module Applica8on Control Module Applica8on Control Plane Northbound API Network Control Module Applica8on Control Interface Network Control Module SDN Network Legacy Network SDN Network Control Plane Control Plane Control Plane Westbound API Eastbound API Southbound API Hypervisor Hypervisor vswitc h Hypervisor vswitc h v User User User Cloud SDN WAN Legacy WAN 4
.Features u Programmability Principle and also key feature of SDN Opens control plane to innovation and enables customization u Protocol independence Compatibility with other networking technologies & protocols Enables technology migration and application-tailored network stacks u Ability to dynamically modify network parameters Active modification of network parameters close to real time Enables fast and flexible adaptation in changing environments u Granularity Control of traffic flows on varying aggregate level and protocol layers Ensures scalability of the control plane to work on different levels u Elasticity Describes the ability of the SDN control plane to scale up and down Enables the control plane to react to variations in traffic mix and volume. 5
and Use Cases u Cloud Orchestration: Provisioning and operation of cloud applications requires integrated management of network and cloud framework u Load Balancing: Integration of load balancing within network forwarding elements operating on different granularities u Routing: Centralized control plane in SDN provides ample opportunities for routing protocol adaptation u Monitoring and Measurement: Ability to perform certain network monitoring operations and measurements without additional overhead u Network Management: Automatic adaptation of network policies based on monitoring information u Application-Awareness: Better cross-layer optimization between applications and network capabilities 6
Use-Cases and Interfaces Use Case Interface Southbound Interface Northbound Interface Eastbound Interface Westbound Interface Cloud Orchestration X X Load Balancing X Routing X Monitoring and Measurement Network Management X X Application- Awareness X X X 7
EXAMPLE: CLOUD ORCHESTRATION 8
Migration Intra DC Energy Op?mizer Cloud Mgmt. Module QoE Op?mizer SDN Network Control Plane VM1 VM2 VM3 9
Migration Intra DC Energy Op?mizer Cloud Mgmt. Module QoE Op?mizer SDN Network Control Plane VM1 VM2 VM3 10
Migration Intra DC Energy Op?mizer Cloud Mgmt. Module QoE Op?mizer SDN Network Control Plane VM1 VM2 VM3 11
Migration Inter DC Network Control Network Mgmt. Module WAN Op?mizer SDN Network Control Plane Energy Op?mizer Cloud Mgmt. Module SDN Network Control Plane QoE Op?mizer Energy Op?mizer Cloud Mgmt. Module SDN Network Control Plane QoE Op?mizer VM1 VM2 VM3 12
Migration Inter DC Network Control Network Mgmt. Module WAN Op?mizer SDN Network Control Plane Energy Op?mizer Cloud Mgmt. Module SDN Network Control Plane QoE Op?mizer Energy Op?mizer Cloud Mgmt. Module SDN Network Control Plane QoE Op?mizer VM1 VM2 VM3 13
Migration Inter DC u Problems: Variability of traffic Application requirements Interaction between controllers etc B4: Software- Defined WAN (Google, ACM Sigcomm 2013) 14
Current Research Topics u Performance evaluation of the SDN architecture u (Controller placement and controller architectures) u (SDN-based application and network interaction) u NFV placement and performance 15
PERFORMANCE OF THE SDN ARCHITECTURE 16
Performance of the SDN Architecture u Performance of the data plane u Performance analysis of SDN Controller u Modeling and performance evaluation of the SDN architecture 17
Performance of the Data Plane u Analysis of throughput and processing delays of OpenFlow enabled forwarding devices Open v NetFPGA Pronto OpenFlow-enabled switch u Testbed to measure data plane performance of devices Link rate of 1Gbit/s Endace DAG card to capture traffic 18
Results Number of Forwarding Rules u Processing delay for a nearly empty (one rule) and a full flow table u Significant impact of payload length on processing delays u High impact of flow table entries on NetFPGA performance 19
Results Forwarding to Controller u Impact on processing delays by forwarding all packets to NOX controller u Massive packet loss between 95% and 99% u Significantly increased processing times u Controller acts as bottleneck in this scenario 20
Performance Analysis of SDN Controllers u Analysis of KPIs of SDN controller software in realistic environments Throughput, latency, CPU & RAM, IAT, Holistic framework for different OpenFlow versions u Implementation of OFCProbe Emulates data plane message and resulting control plane traffic u Features Generated control messages per switch Topology emulation and PCAP file playback Incoming data packets can be arbitrarily distributed 21
Outstanding Packets: Floodlight u Floodlight: Uniform handling of particular switches - consistent behavior u Nox: Non-uniform handling waves detectable 22
Performance Evaluation of SDN u Investigation of the performance of the SDN architecture for changing parameters Modeling of control and data plane System scalability and limitations of the concept u Evaluations using analytical modeling and simulations Input parameters based on measurements with real hardware Verification of analytical results with simulations u Simulation of OpenFlow OpenFlow implementation for OMNeT++: OFOmnet Code available at https://github.com/lsinfo3/ofomnet 23
Simple Model of SDN u Abstraction as feedback-oriented queuing system model Forward queueing system of type M/GI/1 Feedback queueing system M/GI/1-S 24
Results for Different Forwarding Probabilities u Impact of different forwarding probabilities on the average packet sojourn time u Mean sojourn time increases for increasing controller load and for increasing forwarding probability 25
SDN Performance: Summary u Performance analysis of SDN architecture and SDN control plane Controller analysis using OFCProbe Performance evaluation of the architecture using models u Main results of the current investigations Diverse behavior of software control planes, e.g., Floodlight outperforms NOX in terms of throughput and fairness Scalability mainly depends on control plane u Other issues: Investigation of different topologies and software controllers Integration and investigation of OpenFlow 1.3 Impact of messages via Northbound interface Extension of the analytical models 26
NFV PLACEMENT AND PERFORMANCE 27
Network Functions Virtualization 28
NFV in Mobile Networks u Problem: Mobile Core consists of numerous expensive, proprietary, overdimensioned middle boxes. u Idea: Move network function into software (NFV) Run and orchestrate it in cloud u Advantages: Shorter release cycles Elasticity Flexibility Photo: Ericsson u Showcase: Dynamic instantiation of Serving Gateways (SGW) in case of increased resource usage caused by mega events 29
NFV & SDN Cloud Network Management Security Rules SDN Controller Cloud Controller A1 A2 VNF 1 Cloud Infrastructure with virtualized appliances and virtualized network functions External network SDN Legacy SDN Legacy SDN Smartphone Use case : Network Function IT client IT client 30
NFV: Placement and Performance of VNFs u Performance analysis of virtualized network functions u Placement of virtualized network functions (VNFs) 31
SDNA Software Defined NFV Application MEGA EVENT USE CASE 32
Mobile Network Infrastructure Home Ben Ann Event Data center SGW NE+ POCO* NUC* CAM* * CAM Cloud Application Manager NUC Network Utilization Control POCO Pareto-Optimal Resilient Controller SGW Serving Gateway Operator Control Center SGW App SGW Controller 33
Increased Resource Requirements for Mega Events Event Area Ann Event Area Ann Ben Ben Home Area Home Area 34
Planning Infrastructure on Demand 35
Flexible Reuse of Existing Infrastructure Home Ben Event Ann Video call Video call Data center SGW SGW App SGW Controller SGW * CAM Cloud Application Manager NUC Network Utilization Control POCO Pareto-Optimal Resilient Controller SGW Serving Gateway NUC* POCO* CAM* Operator Control Center SGW App SGW Controller 1. Deploy SGW App and Controller à CAM 2. Program virtual GW à SDN+CAM 3. Security check 36
Virtualized Network Functions in Operator Cloud Ø Scalability Ø Redundancy Ø Flexibility Ø Open Source platform CAM NUC 37
PERFORMANCE OF VIRTUALIZED NETWORK FUNCTIONS 38
Performance of Virtualized Network Functions u Impact of softwarization on performance of network functions Impact on typical KPIs, i.e., delay, throughput Influence of dynamic function placement u Categorization and Modeling of VNFs By resource demands: CPU-intense, network-intense, etc. By ability to scale out: scale out delay, state-sync, etc. Identification and investigation of characteristic VNFs Analysis of the influence of the virtualization platform 39
Performance of a Virtualized Firewall u Comparison of Cisco ASA/ASAv in a dedicated testbed Cooperation with the computing center of UniWü u Measurement-based comparison of virtualized and hardware Cisco ASA Firewall Data plane performance (throughput, connection setup) Configuration and monitoring via REST API u Entities under investigation: ASA Service Module (Hardware) ASAv on vmware / KVM Internal network Firewall Module External network 40
Summary u SDN interfaces are key to integration and better user experience Interaction with legacy infrastructure and cloud controller Tailored handling of traffic flows or aggregates Application-aware networking ensures optimal user experience u SDN control plane is performance-critical for the whole network Measurement and simulation tools provided Suggestion of (simple) analytical model Optimal controller placement and hierarchy under investigation u Network Functions Virtualisation (NFV) as logical step, supported by SDN Open issues regarding performance of pure software implementations, interfaces, placement, operations, monitoring,... Benefit: Flexibility of the network as we know it from software Mobile network operators are planning rollout of virtualized EPC 41