Measuring the Mobile Internet David Choffnes Northeastern University with U. Michigan, USC, Google, INRIA, and many others
Motivation Most of the time people spend online is via a mobile device* When YouTube hangs, is it } A slow device? } Weak signal? } Contention for bandwidth? } Bad path to Google? } ISP shaping? } Bad server? Key challenge : We need extensive network measurements to perform root cause analysis Measuring the Mobile Internet 2
There’s an app for that Solution : App-based measurement } Speedometer } Mobiperf } MySpeedTest } Netalyzr Mobile } Namehelp Mobile } … How is this working out for us? Measuring the Mobile Internet 3
Lessons learned from mobile measurement } What you measure may not be what you think Example: Estimate page load time for google.com (low cost) 1. DNS lookup for google.com 2. Ping google.com 3. Estimate latency as DNS lookup time + N RTTs } But when you do an HTTP GET Cellular Link Wired Link for google.com… Non-Split: } Connection is split ① HTTP GET www… Extra DNS lookup } New DNS lookup based on ① HTTP 200 OK Host: field Split: ① HTTP GET www… ② HTTP GET www… } …except for YouTube + T -Mobile ① HTTP 200 OK ② HTTP 200 OK Measuring the Mobile Internet 4
Lessons learned from mobile measurement } Spray and pray measurement can get you breadth… HSPA HSDPA UMTS EDGE GPRS LTE EVDO eHRPD 1xRTT 439K 2326K 563K 506K 58K 1460K 2183K 301K 68K # of Measurements 50 111 96 85 48 7 8 2 3 # of Carriers } …but noise in the data frustrates characterization 1000 GPRS EDGE UMTS HSDPA HSPA Ping RTT (ms) 100 T A Y S V V V V O A T R S N T S S E - T e w o o o o 2 i e o i T e F K m M r n & s d d d d ( t l g T l R i e k g s o O s U T o a a a a o e t T s l T b D r e b f f f f K m r p c o o o o e a i s l l i t ) o e e l o n n n n l e u s C c m e e e e e s o ( ( ( ( l o D N I U m E M E L K ) o ) ) ) Measuring the Mobile Internet 5
Lessons learned from mobile measurement How do we isolate cases of bad performance? [PAM ’14] Example : Path inflation evolution in mobile networks } Combination of traceroute and DNS lookups } Knowledge of Google’s infrastructure } Continuous measurements Example : Shotgun correlations } Vary signal strength only, 0.26 1800 Ping RTT Packet Loss 0.24 1600 observe impact on performance 0.22 0.2 Mean Packet Loss (%) 1400 0.18 1200 0.16 } Doesn’t really generalize to factors 0.14 1000 0.12 800 0.1 0.08 such as location 600 0.06 0.04 400 0 5 10 15 20 25 30 ASU Measuring the Mobile Internet 6
Next generation of mobile measurement Our 1st-generation apps have been useful! Making further strides in this domain } Needs a large(r) set of participating devices } Real-time monitoring of performance } On-demand (adaptive) measurements for RCA Great, let’s all go build new apps! Measuring the Mobile Internet 7
Pitfalls of App Proliferation Coverage } Who has a billion-install idea? Million-install? If so, what the heck are you doing here? } Consistency } How do we join datasets? } What measurement implementations? Cost } How many times do we need to run the same ping? Interference } What if MST runs during Mobiperf’s ping test? Measuring the Mobile Internet 8
This PowerPoint 2007 template produces a 36”x48” Modify scheduler’s configuration Mobilyzer: Mobile Measurement Library Ping,“google.com”, click on “Layout” to see Our proposal : One measurement library, many apps } API for measurements simplifies app dev the “Design” menu and click on “Colors”. You can Measuring the Mobile Internet 9
This PowerPoint 2007 template produces a 36”x48” Modify scheduler’s configuration Mobilyzer: Mobile Measurement Library Ping,“google.com”, click on “Layout” to see Our proposal : One measurement library, many apps } API for measurements simplifies app dev } Single scheduler per device allows apps to coexist the “Design” menu and click on “Colors”. You can Measuring the Mobile Internet 10
This PowerPoint 2007 template produces a 36”x48” Modify scheduler’s configuration Mobilyzer: Mobile Measurement Library Ping,“google.com”, click on “Layout” to see Our proposal : One measurement library, many apps } API for measurements simplifies app dev } Single scheduler per device allows apps to coexist the “Design” menu and click on “Colors”. You can } Data collection Measuring the Mobile Internet 11
Why Mobilyzer? Designed with researcher incentives in mind } Simplifies app development } Just ask MySpeedTest dev (GaTech student) } PlanetLab-like participation model } Bring one Mobilyzed app, run measurements on all devices } Dynamic server-side control of measurements Measuring the Mobile Internet 12
One last thing What about detecting traffic differentiation in mobile? Measuring the Mobile Internet 13
One last thing What about detecting traffic differentiation in mobile? } Challenges } What traffic triggers differentiation? } How do we tell that the ISP is doing anything (vs. noise)? } Our approach } Record and replay traces of real traffic from mobile apps } Use VPN tunnels ( Meddle ) as controlled experiments App to be released soon (works without root)! Measuring the Mobile Internet 14
Thanks! Obligatory project URLS http://mobilyzer-project.mobi http://mobiperf.com http://meddle.mobi http://openmobiledata.appspot.com Measuring the Mobile Internet 15
Recommend
More recommend
