characterizing home wireless performance: the gateway view ioannis pefkianakis* h. lundgren^, a....
TRANSCRIPT
Characterizing Home Wireless Performance: The Gateway View
Ioannis Pefkianakis*
H. Lundgren^, A. Soule^, J. Chandrashekar^, P. Guyadec^, C. Diot^, M. May^, K. Doorselaer^, K. Oost^
HP Labs* Technicolor^
Today’s Residential WLANs
Wireless gateway
WiFi repeater
Tablet
Wireless baby monitor
Wireless gateway
Laptop
Smartphone
Microwave oven
Multitude of Wi-Fi devices running high-bandwidth apps
ISPs Strive to Understand Wi-Fi Home Nets
• What and how many devices dominate the traffic?
• What is the wireless performance?
• How often do users experience poor performance?
• What is behind poor performance?
Existing Approaches on Understanding Home Nets’ Performance
• The customized-AP approach allows for fine time scale measurements from all the devices connected to the AP
– [WiSe, MOBICOM’13], [Papagiannaki et al., INFOCOM’13], [BISMark]
– Small-scale deployments of technically inclined volunteers
• The end-host measurement tools run as apps and collect feedback at the client side
– [Home Net Profiler, PAM’13]– One shot measurements, limited application-level feedback
• Our approach: Collect data from the home gateways of the subscribers of a large ISP under normal service operation
Outline
• Measurement infrastructure and deployment
• Metrics
• Wi-Fi environment and traffic dynamics
• Wireless performance
• Root cause of performance bottlenecks
Measurement Infrastructure
Broadband Network
Dashboard
ControllerData storage
WiFi repeater
Tablet
Wireless baby monitor
802.11b/g/n wireless gateway
OSGI Bundle
Passive measurements from subscribers’ gateways
Why Home Gateways?
• Gateways offer a complete view of the home network
– Continuously monitor all the devices connected to the gateway
– Observe neighboring Wi-Fi networks
– Capture both wireless link performance and traffic dynamics
• Using existing infrastructure allows for large-scale, more diverse deployment
Dataset
• 167 gateways (71% fiber, 29% ADSL) in 10 different cities
• gateways report every 30 seconds
• 4-month (June-September 2013) collection campaign
• 1328 Wi-Fi devices detected
Metrics
• What we have– PHY rate
• Performance indicator
– RSSI • Wireless coverage metric
– Traffic counters
– Neighboring SSIDs and their RSSI’s
• What we miss– Frame losses
– Channel contention
– We cannot capture the actual wireless throughput
How to Capture Wireless Problems?
• Coverage • Interference
Wireless gateway Tablet
weak signal(low RSSI)
Map RSSI to Speed(in RF chamber)
RSSI (dBm) Expected PHY rate (RE)
[min, -88] 6.5 Mbps
… …
[-70, max] 65 Mbps
Wireless gateway TabletWireless
baby monitor
loss
PHY rate R dropsbut RSSI remains the same
RateGap = Rate_index(RE)-Rate_index(R)
Metrics: Putting Everything Together
High RE Low RE
High R good performance poor performance(RA dynamics)
Low R poor performance(interference/RA dynamics)
poor performance(poor coverage)
Outline
• Measurement infrastructure and deployment
• Metrics
• Wi-Fi environment and traffic dynamics
• Wireless performance
• Root cause of performance bottlenecks
Wi-Fi Environment
• High penetration of the newer 802.11n devices
– 0.5% .11b, 42.5% .11g, 45.5% .11n 1x1, 11.5% .11n 2x2
• Diversity in the number of Wi-Fi home devices (1 to 25)
– Median home has 4 resident devices (i.e., observed for several days)
Wi-Fi Traffic Dynamics
Traffic is generated by a few devices3 Wi-Fi devices generated the most traffic in 70% of the homes
… during evening times
Outline
• Measurement infrastructure and deployment
• Metrics
• Wi-Fi environment and traffic dynamics
• Wireless performance
• Root cause of performance bottlenecks
16
What is Home Wireless Performance?
Wireless link performance (i.e., PHY rate) is overall good!
Effective (f(PHY Rate, 802.11 overheads)) higher than actual throughput
… but there are still performance bottlenecks (for 7.6% of the samples PHY rate <=6.5Mbps)
throughput gap > 20Mbps for most of the homes
Performance Variation Across Homes
The fraction of poor performance episodes varies across homes
for most of the homes poor episodes are ≤ 6%
Poor performance episodes can be up to 66%
18
Root Cause of Poor Performance
Metric: Convert RSSI signals to an expected link speed (PHY Rate)
Wireless coverage is not likely a cause of poor performance
78% of the transmissions
at the peak expected PHY rate
Root Cause of Poor Performance
Poor performance can be caused by interference and PHY rate adaptation dynamics
High RateGap can lead to poor performance
for 18% of the instances RateGap>4
RateGap varies across homesThe peak RateGap corresponds to the 2 homes
with the highest poor performance instances
Interference Causes
• Contention from in-home Wi-Fi devices is low– For the majority of homes (78%) local contention is less than 10%– Interference can be attributed to external sources (non-Wi-Fi devices,
neighboring Wi-Fi networks)
• There is no strong correlation between Wi-Fi performance and the density of the neighboring Wi-Fi environment.
Conclusion
• We study Wi-Fi home networks of the subscribers of a large ISP
• Wireless link performance (i.e., PHY rate) is overall good
• We still identify instances of poor performance, where we eliminate poor coverage to be their root cause
• ISPs’ helpdesk calls for wireless problems may not be attributed to the wireless link
– … but to gateway misconfigurations, authentication problems, end-device issues
Thank you!