SELF MANAGEMENT OF COGNITIVE FUTURE INTERNET ELEMENTS Presenter : Nancy Alonistioti (nancy@di.uoa.gr ) National and Kapodistrian University of Athens
SELF MANAGEMENT Future Internet A complex adaptive organization, where the involved partners have conflicting goals and tension to maximize their gains There is a need for new ways to organize, control and federate communication systems How to govern the increasing autonomic behaviours in hybrid communication systems? Cognitive Cycle Cognitive capabilities enable the perception of the network elements environment and the decision upon the necessary action (e.g. configuration, healing, protection measures etc.) The feedback control cycle model, (Monitoring Decision Making Execution) is expected to be in the heart of Future Internet Elements Network elements with cognitive capabilities aim at fast localised decisions and configuration actions as well as learning capabilities that improve their behavior 2
FUNCTIONAL ARCHITECTURE FI cognitive managers Network Domain Cognitive Manager Network Element Cognitive Manager Operator Policy Repository Management Level Cognitive Level Domain B Domain A BTS NodeB Router Router Router Router Router Router NodeB Router NodeB BTS Networking Level 3
Defining autonomic behaviours The question is to be able to specify system behaviours combining Autonomous feedback control loops And overall scalability and stability of the system Self management perspective and contributions: Contain low level control loops in a specific domain (LAN, small number of cells in the 3G network, ) Forward chaining for taking actions + learning loop for S3 parameter tuning S1 S2 S4 S5 S6 S7 C14 C2 C3 C4
Goals: EXPERIMENTATION OF SELF MANAGEMENT CAPABILITIES: USAGE OF PANLAB FIRE FACILITY 1. Use case : Service adaptation and network reconfiguration based on multi objective optimisation (e.g., QoS, packet loss, fault, interference etc.) 2. Use Self NET prototype results as a basis. Address experimentation based on the use in diverse platforms and larger scale. 3. Address experimentation for autonomic communications (internal Self NET results and external from FIRE facilities) Steps: 1. Agree on the use case and the testbed features from N.K.UoA and Octopus side 2. Build the IPIP Tunneling for the federation of the Self NET and Panlab facilities 3. Run manually initial experiments for triggering events and configuration actions testing 4. Build NECM/NDCM based on PANLAB facility features for the automation of experiments from N.K.UoA side 5. PII develop Resource Adapters in order to automate testbed resources reservation 6. Repeat Experiments 5
PANLAB TOPOLOGY FEATURES Airspan MicroMAX WiMAX base station and subscriber stations is located on the Octopus testbed at Oulu (VTT). WiMAX operates in FDD mode using 3.5 MHz bandwidth for the DL and UL at 3.5GHz. The user traffic from the Self NET environment is tunnelled using two IPIP tunnels over the Internet and rerouted over the WiMAX air interface at the Octopus testbed Distributed Internet Traffic Generator (D ITG) Self NET has implemented the Network Domain Cognitive Manager (NDCM) as well as the Network Element Cognitive Manager (NECM) Network Level (i.e. WiMAX BS) NECM is used for the WiMAX BS monitoring (though SNMP get) and configuration actions (web services API/SNMP GET) Service Level NECM is used for the Service level monitoring and configuration actions NDCM software is associated with the NECM entities 6
OCTOPUS TESTBED FACILITIES AND SELF NET SOFTWARE FEDERATION 7
1 U NETWORK TOPOLOGY AND IPIP TUNNELLING 1 U 8
USE CASE DESCRIPTION 1. Various types of traffic (Data, VoIP, Video) are introduced to the BS from the wired interface (DownLink for associated terminals to WiMAX BS) and also through the associated clients (UpLink for associated terminals to WiMAX BS) 2. BS NECM: Measure through SNMP GET (using web services), metrics from WiMAX Base station: UL/DL used capacity, TCP/UDP parameters, number of flows 3. Service Level NECM: Retrieves associated clients perceived QoS (packet loss, delay, jitter) and the features of the service (VoIP, ftp, video) that service provider(s) offer 4. NDCM: According to the collected monitoring information (step 2 and step3) the NDCM will 1. Identify faults or optimization opportunities (e.g., high packet loss) 2. Service level (i.e. traffic generator) adaptation e.g., change data rate, codec of VoIP or Network level adaptation e.g., change prioritization is decided 5. Execution a) WIMAX BS(network level): Change Prioritization to x number of flows (high, low) b) Traffic Generator (Service level): Change the codec of x flows PANLAB Self NET USE CASE 9
SERVICE FLOWS Several types of VoIP Codecs have been used G.711.1 (48 kbps), G.711.2 (40 kbps), G.729.3 (8 kbps), G.729.2 (7 kbps), G.723.1 (5 kbps) TCP and UDP flows are used as background traffic Different number of VoIP flows Each codec is tested on both High and Low priority @ WiMAX BS side The port of the service flow is identify for the high or low priority High Priority ports: Port range [9850, 10100] Low Priority ports: Port range [10101, 10250] PANLAB Self NET USE CASE 10
DEPLOYED NETWORK AND SERVICE MANAGEMENT SCHEME Monitor Service Level (Service Features, Client QoS) NECM M Monitor Network Level (WiMAX BS SNMP GET) NDCM D Situation Deduction Identification of High PER status, High Delay Change Prioritization of n flows at the WiMAX BS and Video Codec of k flows Decision Making Change Prioritization of n flows at the WiMAX BS Change Video Codec of k flows Change Flows Codec (Service-Level Adaptation) NECM E Change Flows Priority (Network-Level Adaptation) PANLAB Self NET USE CASE 11
DECISION MAKING ALGORITHM FOR CONFIGURATION ACTION SELECTION Different type of VoIP flows traverse the network Possible Actions: Change ALL the flows(ni) of Ci Change some flows of all codecs (xi/ni), for each i Change some flows of some codecs (xi/ni), for specific is' 12
DECISION MAKING ALGORITHM FOR CONFIGURATION ACTION SELECTION N i Flows of codec C i, i={g.711.1, G.711.2, G.729.2, G.729.3, G.723.1} G.711.1 (48 kbps), G.711.2 (40 kbps), G.729.3 (8 kbps), G.729.2 (7 kbps), G.723.1 (5 kbps) with weight W i W G.711.1 = 48/48 = 1 W G.711.2 = 40/48 = 0.833 W G.729.2 = 8/48 = 0.166 W G.729.3 = 7/48 = 0.145 W G.723.1 = 5/48 = 0.104 5 N TOT = Σ N i i=1 5 Service Cost = (N i * W i ) i=1 Class A Class B 13
DECISION MAKING ALGORITHM FOR CONFIGURATION ACTION SELECTION 5 Service Cost = (N i * W i ) i=1 Marginal_Cost = f (Service_Cost) 20, 0 Service Cost<23 10, 23 Service Cost<27 5, 27 Service Cost<33 Marginal Cost = 1, 33 Service Cost<40 10, 40 Service Cost<50 15, 50 Service Cost<70 30, 70 Service Cost<90 5 Adapted_Number_Of_Flows = Marginal_Cost (N i * W i ) i=2 Experimental Knolwedge 14
PERFORMANCE RESULTS 1/2 Improvement on specific QoS features after the re-configuration actions due to an increase of the packet loss rate of VoIP traffic. Various Codecs low Priority Various Codecs High Priority Number of flows 40 40 Total packets 29704 29549 Average delay 7.141964 s 7.164891 s Average jitter 0.015584 s 0.015977 s Average bitrate 2546.360118 Kbit/s 2420.837946Kbit/s Average packet rate 2777.616579 pkt/s 2454.385064 pkt/s G.711.1: 19 G.711.2: 4 G.729.2: 5 G.729.3: 4 G.723.1: 6 Packets dropped 2.77 % 2.29 % The change of the VoIP codec between the service provider and the end user, selecting the G.729.3 codec, reduces the number of the dropped packets (PER 0.042 %) 15
PERFORMANCE RESULTS 1/2 Improvement on specific QoS features after the re-configuration actions due to an increase of the packet loss rate of VoIP traffic. Various Codecs Low Priority Various Codecs Low Priority Number of flows 48 48 Total packets 23004 23817 Average delay 7.409 s 7.429 s Average jitter 0.019809 s 0.0217 s Average bitrate 2546.360118 Kbit/s 2420.837946Kbit/s Average packet rate 2262.653902 pkt/s 1892.468659 pkt/s G.711.1: 18 G.711.2: 14 G.729.2: 5 G.729.3: 4 G.723.1: 6 G.711.1: 0 G.711.2: 32 G.729.2: 5 G.729.3: 4 G.723.1: 6 Packets dropped 25.65 % 0.042 % Reduction of the packet loss rate after the change of the prioritization at the WiMAX BS From low priority to high priority service class 16
PII & SELF NET VIDEO The video of the PII and Panlab Demonstration is available online... http://scan.di.uoa.gr/prototypes/self-net-and-panlab-demo-12-2010 17
EXPERIMENTATION IN SELF MANAGEMENT: LESSONS LEARNED Issues/Recommendations: Configuration capabilities (e.g., tunneling, service deployment, codecs etc.) Interfaces for interacting with the experimental resources Overhead in terms of effort from both sides for the experimentation and use case deployment minimise learning curve Facility that clearly targets/supports Autonomic Communications experimentation (e.g., support reconfiguration (real time) in various layers) Added value: Diversity of technologies and infrastructures Large scale experimentation capability Advanced experimentation results based on multiple metrics Justification of research results in close to real life conditions Building of technical know how at various domains 18
THANK YOU FOR YOUR ATTENTION!!! 19