Classification of Access Network Types: Ethernet, Wireless LAN, ADSL, Cable Modem or Dialup?: Classification of Access Network Types: Ethernet, Wireless LAN, ADSL, Cable Modem or Dialup? Wei Wei, Bing Wang, Chun Zhang, Jim Kurose, Don Towsley {weiwei, bing, czhang, kurose, towsley}@cs.umass.edu


Outline: Outline Motivation Background Classification scheme Analytical basis for classification Validation Conclusions and future work


Motivation: Motivation Ethernet, wireless LAN, ADSL, cable modem and dialup common access networks dramatically different characteristics Fast and accurate classification of access network type using end-to-end approach is useful constructing application layer multicast tree…


Background: Overview: Background: Overview IEEE 802.11b (WLAN) Shared media Random backoff DOCSIS (Cable Modem) Downstream: broadcast Upstream: shared using request/grant Ethernet, ADSL and dialup Essentially dedicated access


Background: Transmission Overhead of WLAN: Background: Transmission Overhead of WLAN Fixed overhead of UDP packet 500 microsecond Random backoff 0 - 620 (perfect channel, no contention) Overall transmission time of UDP packet Uniformly distributed 500 ~ 1120 (perfect channel, no contention)


Classification Approach: Classification Approach A wants to know B’s access network type Let B report its connection type? B may not know (WLAN+cable) B may not want to report A asks B to send a sequence of packet pairs A determines B's connection type based on median and entropy of inter-arrival times of packet pairs from B Assume A has good network connection


Notation: Notation B A Request packet pairs


Classification Scheme: Classification Scheme


Determining Classification Thresholds: Determining Classification Thresholds Independent M/D/1 queues Equal packet size: 375 bytes Average packet size over Internet: 300~400 bytes Single bottleneck Two bottlenecks


Analytical Basis: Analytical Basis 11 Mbps 802.11b, under ideal conditions no contention , no retransmissions Ethernet


Small Scale Experiments: Small Scale Experiments Purpose Validate analytical results Obtain empirical results 12 Linux machines located in 4 continents US, Brazil, Italy, and Taiwan Run experiments between each pair of machines Connection types tested: Ethernet, WLAN, ADSL, Cable


Small Scale Experiments: Small Scale Experiments Ethernet WLAN Cable DSL


Large Scale Experiments: Coverage: Large Scale Experiments: Coverage 28 states in United State Receivers: two PCs in UMASS with 10Mbps and 100Mbps Ethernet card 4 continents, 10 countries Senders are from


Large Scale Experiments: Classification: Large Scale Experiments: Classification Entropy of Ethernet and WLAN can overlap Cable and dialup can have very small median Hard to distinguish cable, ADSL and dialup


Large Scale Experiments: A Dialup Example: Large Scale Experiments: A Dialup Example Median is not enough! A dialup trace from University of Florida looks like an Ethernet connection Need Entropy values for both bin sizes


Large Scale Experiments: Summary: Large Scale Experiments: Summary 509 traces, our classification scheme: correctly classifies all 144 Ethernet connections correctly classifies all 93 WLAN connections as WLAN Low-bandwidth classifies > 98% connections correctly misclassifies < 2% as WLAN All 5 traces are ADSL connections from Calgary, Canada


Conclusion and Future Work: Conclusion and Future Work Simple, efficient scheme to classify access network type: Ethernet, WLAN,low-bandwidth connection exploit intrinsic characteristics of connection types use both median and entropy of packet pair inter-arrival times accurate classification results in 10 - 100 seconds Future work Passive-measurement-based classification Classification of cable and ADSL based on more specific characteristics


Slide18: Thank you!