OpenDaylight Based Machine Learning
for Networks
Contributions from the ODL MILA WorkGroup
YuLing Chen, Cisco
Prem Sankar, Ericsson
Sept. 2016
Contributed by…
Why we need machine learning in ODL?
Why we need Machine Learning in SDN
• Software Defined Network needs to be intelligent.
To be aware of the runtime status of the network.
To make the right decisions to adjust the policies for
the traffic control.
To dynamically change the policies according to the
analytics results.
Make ODL the Brain of Network
VPN
service
GBP
Topo
TSDR
AI/ML
OF/Netconf/BGP
SFC
Use Cases of a smart and intelligent SDN controller
 Traffic Control and Routing Optimization
• Congestion Control
• Traffic Pattern Prediction
• Routing Optimization
Resource optimization
• Networking resource allocation optimization
• Cloud resource management optimization
 Security and Anomaly Detection
• DDoS Attack detection
 Troubleshooting and Self-healing
Example Use Case – Traffic congestion prediction with
automated control
Prediction using Weka leveraging data collected in TSDR
#ODSummit
The Goals of ODL MILA Group
Goals of ODL MILA Workgroup
• To adapt SDN architecture to machine learning requirements.
– To provide an application framework for Machine Learning in ODL
– To add the necessary components for Intelligence(Knowledge) Plane
– To integrate with ODL native network data collection and traffic
control services.
• To facilitate machine learning application development on ODL.
- Integrate with third party machine learning algorithms
- Provide abstract and generic northbound interfaces for Machine
Learning applications
- Hide the details of Advanced Analytics and Machine Learning
complexities.
How to realize a smart and intelligent SDN
Controller
• Network status awareness
 Rely on time series data
collected from the network
Advanced Analytics
& Machine Learning
Automated Traffic
Control
Time Series Data
Collection
• Traffic Control Policy Change
decision making
 Based on the advanced
analytics and machine
learning.
• Dynamic change of Control
policies
 Automatically change the
traffic control policies based
on the analytics results.
Time Series Data Repository
TSDR Capabilities and Architecture Framework Roadmap
Control Flow Data Flow
TSDR Integrated Architecture in ODL
 TSDR Data Services including
Data Collection, Data Storage,
Data Query, Data Purging, and
Data Aggregation are MD-SAL
services.
 Data Collection service receives
time series data published on
MD-SAL from MD-SAL
southbound plugins.
 Data Collection service
communicates with Data Storage
service to store the data into
TSDR.
 TSDR data services access TSDR
Data Stores such as HBase Data
Store through generic TSDR Data
Persistence Layer.
ODL MILA Framework Architecture
ODL MILA framework in the ODL ecosystem
•
•
•
•
•
Enable AI/ML on both historical and
real-time data paths.
Many use cases would require both
offline and online ML on the time
series data.
External events could be additional
input for accurate machine learning
results.
Feed back the results to SDN control
path for automatic traffic steering and
policy placement.
Well-defined interface among the
components towards future
standardization of advanced analytics
in SDN.
ODL MILA framework PoC Architecture
• PoC of both historical
offline machine learning
and real-time online
machine learning
 Collect the time series
data
 Persist into scalable data
storage
 Publish to high
performance data bus
• Integrate with external
machine learning libraries
 Spark MLlib
 DeepLearning4J
• Collect OpenFlow Stats
and apply machine
learning algorithms
 k-means clustering
ODL MILA framework PoC Result
• Snapshots of flow size
categorization are captured
using k-means clustering
algorithms.
 Different colors show the
different category/cluster that
each flow falls into based on
the flow size.
Demo
ODL MILA Workgroup related info and contacts
• https://wiki.opendaylight.org/view/Useful_Links_and_Learning_materials
• Contacts
• YuLing Chen [email protected]
• Prem Sankar [email protected]
Thank You