Monitoring and MiningNetwork Traffic in Clouds Edward J. Yoon <email@example.com>
Contents• Introduction• Analysis for Target Marketing – Overview of Proposed Methodology – Network Usage Trend Analysis – Customer Similarity Analysis – Abnormal Traffic Analysis• Determining the System Expansions
IntroductionWhy do we need monitoring and mining the network traffics?• Business points: – Target marketing – Detecting abnormal traffic• System management points: – Determining the system expansions – Network planning and design
Analysis for Target MarketingTraffic is most important factor of finding ideal target customer:• Network Usage Trend Analysis • who grows faster in particular time period?• Customer Similarity Analysis • who shows a similar pattern of evolution?
Overview of Proposed System - Network Usage Trend Analysis - Customer Similarity Analysis - Abnormal Traffic Analysis - … Analysis Map/Reduce Hama BSP NetFlow Database on Hadoop
Network Usage Trend Analysis• Methodology: Traffic Growth Rate (value at end of period - value at beginning of period) PR *100 value at beginning of period
Customer Similarity Analysis• Methodology: Vector Space Model – Growth Rate Vector (or, Payload Vector, ...) – Cosine Similarity • The most common similarity metric in NLP
Abnormal Traffic Analysis• Methodology: Entropy, Chi-Square Test, .. – Entropy Analysis • E.g., measure of source IP disorder n H Pi log2 Pi i 1
Determining the System Expansions• System Expansions Planning• Network Planning and Design• .. and so on.
What’s Next?• VM re-placement based on Resource Usage Pattern.VM1 VM9 … VM2 VM4VM2 VM5 VM7VM3 VM1 VM3 VM2 Replacement t