Technologien und Mobilkommunikation Self-Healing in Self-Organising - PowerPoint PPT Presentation

Lehrstuhl Netzarchitekturen und Netzdienste Institut für Informatik Technische Universität München Seminar Innovative Internet Technologien und Mobilkommunikation Self-Healing in Self-Organising Networks Oliver Scheit

Self-Organising Networks  Self-Organising Networks in Mobile Communications  Motivation  Self-Configuration  Self-Optimization  Self-Protection  Self-Healing  Detection  Diagnosis  Problems Self-Healing in Self-Organising Networks 2

Motivation  Tendencies from macro to micro/pico/femto cells in networks  To overcome the bandwidth problems  More and smaller cells needed for coverage  Paradigm change for management purposes: • Network configuration: – Move from centralized planning and configuration to decentralized Self Organizing Networks • Network maintenance and optimization: – Cells autonomously gather information and provide it to a management system – This cannot be done manually by personnel any longer Self-Healing in Self-Organising Networks 3

Motivation  Heterogenous networks with GSM, HSPA, LTE  Interference  Knowledge of the network often only by experience  Dependency on individual experts  Typical operater use cases:  Planning  Deployment  Optimisation  Maintenance  Find a cost efficient way to manage the network tasks Self-Healing in Self-Organising Networks 4

Self-Management Domains • Configuration • Resource of network monitoring parameters • Parameter tuning • Network planning Self- Self- Configurati Optimisatio on n Self- Self-Healing Protection • Protection • Error detection against attacks and recovery • Faulty operation Self-Healing in Self-Organising Networks 5

Self-Configuration  System can add and configure new elements/features in run-time  Reduce time until first operation  Improves deployment of new network elements  Automatically install new software  Reduce human involvement Self-Healing in Self-Organising Networks 6

Self-Optimisation  Improve efficiency of the network over manual configuration  Adapting to changing environment  Handover parameters, cell-individual parameters  Neighbor cell list  Monitor the system parameters  Automated parameter tuning  Mobility Load-Balancing Self-Healing in Self-Organising Networks 7

Self-Healing  Detection  Sleeping cells  Performance indicators  Diagnosis  Use network knowledge  Prevent false alarms  Recover/reduce impact on network performance  Cell Outage Compensation  Reconfigure neighbor cells to compensate cell outage Self-Healing in Self-Organising Networks 8

Network Architecture Network NM Management Domain DM DM Management Network NE NE NE Element Where to implement which SON function ? Self-Healing in Self-Organising Networks 9

Challenges of SON-Architecture  Centralised vs distributed functions  Processing power vs bandwidth  Centralised Decision making (more knowledge) vs distributed (deadlocks, scaling)  Reliability & availability  Management Self-Healing in Self-Organising Networks 10

Self-Healing  Mobile networks are large, complicated systems  Prone to faults and inefficient behaviour  Most critical part is the Radio Access Network  Little to no redundancy in base stations  No service for users if one station fails completely  High amount of network elements  Degradation possible in each element  Identification of root cause for triggered alarms is hard  Often requires manual troubleshooting  Long time period of degradation  Significant costs Self-Healing in Self-Organising Networks 11

3GPP Self-Healing  Use-cases derived by 3rd Generation Partnership Project (3GPP)  Self-recovery of NE-software  Self-healing of board faults  Cell Outage Detection  Cell Outage Recovery  Cell Outage Compensation  Return of Cell Outage Compensation Self-Healing in Self-Organising Networks 12

3GPP Self-Healing Process  Self-Healing process  Input monitoring checks for pre-defined conditions, triggers self-healing process  Self-healing process gathers additional system information  Diagnosis of the root cause with the provided information  If root cause can be solved automatically: • recovery action performed • Back up configuration data  Evalution of the new state • Attempt new self-healing iteration • Report the result of the self-healing process • Fall back to backup configuration Self-Healing in Self-Organising Networks 13

Cell Degradation Management  Detection of Cell degradations partly automated  Check for alarms and faults  Reset the cell  After too many resets, operational personnel investigates the cell  If it can‘t be fixed remotely, engineers investigate on site  Often multiple visits requiered  KPI based detection  Investigate key performance indicators (KPI) • E.g. Top 10 cells with dropped calls, change of dropped call rate  KPIs often indicate external influences like interference  Faults also typically related to hardware Self-Healing in Self-Organising Networks 14

Cell Degradation Detection  Detecting unusual behavior of performance indicator  No diagnosis yet  Decide if performance is „healthy“  Performance indicators aquired from  Base station  User equipment  Neighbor cells  Core network elements  KPIs are stochastic variables  Absolute decision what is „healthy“ or not impractical  Use profiles to evaluate operation Self-Healing in Self-Organising Networks 15

Profiles  Absolute Threshold  Indicator should not exceed or fall below certain threshold [1] Figure 1: absolute Threshold Self-Healing in Self-Organising Networks 16

Profiles  Statistical  Values that should be around the statistical mean of the indicater +/- a threshold defined by the operator [2] Figure 2: statistical thresholds Self-Healing in Self-Organising Networks 17

Profiles  Time dependend  Indicators fluctuate time dependend, usually these are dependend of user behaviors [3] Figure 3: absolute Threshold Self-Healing in Self-Organising Networks 18

Cell Degradation Detection  The performance of the Detection method can be evaluated by  The delay between a degradation and its detection  The accurancy false postive/false negative  The processing Overhead Self-Healing in Self-Organising Networks 19

Cell Degradation Diagnosis  Diagnosis usually done manually  Requires knowledge of typical symptom-cause relation  This is often refered to as expert knowledge  SONs have to map the observations to the most possible root causes similar to a way a human solves this problem  different approaches to this problem possible  Rule based systems  Bayesian Networks  Case based reasoning  Neural Networks Self-Healing in Self-Organising Networks 20

Rule Based Systems  If(a)-then(b) style of implementation  Easy to implement  Easy to train/add new rules  Only effective in small deterministic networks  Complex networks require a complex set of rule  Uncertainty to performance indicators Self-Healing in Self-Organising Networks 21

Bayesian Networks  Includes the uncertainty of the performance indicators into the analysis of the root cause  Based on Bayes Theorem P(A|B) = (P(B|A)*P(A)) / P(B)  Can be depicted as an acyclic graph with nodes on different hierarchies  Conditions (global properties)  Root causes  Sypmptoms  Weighted edges represent the probability that a fault causes an observed symptom  Can be build top down with knowledge of the operator  Can evolve over time, starts with limited knowledge Self-Healing in Self-Organising Networks 22

Bayesian Networks Condition Root Root Cause Cause Symptom Symptom Symptom  Faults are rare, their statistical distribution hard to define  Size of conditional probability table grows exponentially with the number of nodes in the network Self-Healing in Self-Organising Networks 23

Case Based Reasoning  Map KPI anomalies to Fault cases  Initially requires expert knowledge  To solve a new problem, knowledge based on the similarity to other already known faults  If the fault could be resolved, add this new fault with the corresponding KPI mapping to the database  If it can‘t be solved, manual troubleshooting is done  The new case will be added to the knowledge base  CBR does not rely on probability Self-Healing in Self-Organising Networks 24

Conclusion  SON are an efficient way to deal with increasingly complex networks  SON Domains cover deployment, configuration and self- healing of network elements  The SON Task can be done centralized by an higher Network Management element or decentralized by the cells themselves  Detection is linked to alarms and performance indicators  Dependent on probabilities  Finding the root cause of a problem is a machine learning problem Self-Healing in Self-Organising Networks 25