understanding user experience on mobile devices with the ...narseo/nanog_maryland.pdf• native...

29
Understanding user experience on mobile devices with the ICSI Netalyzr Narseo Vallina-Rodriguez, Srikanth Sundaresan, Christian Kreibich, Nicholas Weaver and Vern Paxson ICSI-UC Berkeley Moritz Steiner Akamai 1 Opinions are those of the presenter, and not of the funders Project sponsored by:

Upload: others

Post on 09-Aug-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Understanding user experience on mobile devices with the ICSI NetalyzrNarseo Vallina-Rodriguez, Srikanth Sundaresan, Christian Kreibich, Nicholas Weaver and Vern Paxson!ICSI-UC Berkeley!!Moritz Steiner !Akamai

1

Opinions are those of the presenter, and not of the funders

Project sponsored by:

Page 2: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Where I’m coming from

2

• ICSI: International Computer Science Institute - Berkeley

• PhD in Computer Science at the University of Cambridge (UK, 2013) in collaboration with Telefonica Research. Research on mobile OSs and cellular networks!

• Member of ICSI-Networking and Security group since July 2013

• More info:

• Personal page: http://www.icsi.berkeley.edu/~narseo

• Group page: http://www.icsi.berkeley.edu/icsi/groups/networking

Page 3: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

3

The Internet is becoming mobile

Page 4: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Where is the actual bottleneck in mobile

networks?

4

Page 5: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Carte Figurative des Pertes Successives en hommes de l’Armée Francaise dans la Campagne de Russie (1812-1813)

!Charles Minard.1869

5

End-to-end model?

Page 6: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

What is the cause of actual performance?

6

Page 7: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

The ICSI Netalyzr for Android

7

Page 8: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

• User-run tool to diagnose connectivity characteristics from the edge of the Internet

• Checks for behavioral anomalies, network properties and security issues for everybody

• Started in 2009 as a JAVA applet (1.4M sessions in 170 countries)

• Native Android version since November 2013 (22K installs in 120 countries)

• Research outcome: IMC’10, FOCI’11, SATIN’11, PAM’12/13, CONEXT’14

8

The ICSI Netalyzr Story

Page 9: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

IP CORE (USER PLANE)!!

DNS RESOLVERS HIDDEN PROXIES

NETWORK TOPOLOGY AND PEERING SHARED INFRASTRUCTURE

TLS INTERCEPTION IPV6 AND DNSSEC SUPPORT

NATs DETAILED LATENCY TESTS

HANDSET HW AND MISCONFIGURATION!!

APN CONFIGURATIONS BUFFERS

CERTIFICATE ROOTSTORE SYSTEM CLOCK

Netalyzr for AndroidRADIO TECHNOLOGY (CONTROL PLANE)!

9

Page 10: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

!1. Middleboxes deployment and behavior!!2. IPv6 support!!3. Layer 8: Infrastructure sharing and mobile

virtual network operators (MVNO)

The analysis: !Looking beyond the radio link!

10

Page 11: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

!!!!!!!!!

DATA PLANE

The IP core of cellular networks

11

UE!!!!

CONTROL PLANE

NODE B!/ RNC

GGSN!/P-GW

SGSN!/S-GW

INTERNETIP COREN!A!T

DNS !Resolver

PROXY!/CachePDP Context !

(APN configuration)

Page 12: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

The impact of network topology

Inappropriate network deployment can double DNS lookup time!

12

3G

4G

Median DNS Lookup Time (ms)

0 70 140 210 280

DNS Before CGNDNS After CGN

Page 13: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

DNS ProxyTCP:21 (FTP)TCP:22 (SSH)

TCP:25 (SMTP)TCP:80 (HTTP)

TCP:110 (POP3)TCP:143 (IMAP)

TCP:443 (HTTPS)TCP:587 (SMTP)

TCP:993 (TLS IMAP)TCP:995 (TLS POP3)

TCP:1194 (VPN)TCP:1723 (PPTP VPN)

TCP:5060 (SIP)TCP:6881 (TORRENT)

TCP:9001 (TOR)

Percentage of operators

0 20 40 60 80

WorldwideUSA

Hidden proxies on cellular networks

Page 14: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Headermodification

Contentmodification

Header (*)forwarding

HTTP header enforcement

Non-HTTP traffic filters

Percentage of operators

0 17.5 35 52.5 70

WorldwideUSA

What are HTTP proxies actually doing?

Page 15: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Questions for you …

15

• 11% of sessions on IPv6. It requires handset and network support (APN settings)

!• Only 9 operators* in 6 countries: US, Canada, Poland, Italy,

Australia, and Sweden !

• No significant latency difference with IPv4 observed yet !

!

How is IPv6 transition done in practice and what issues does it cause?

Page 16: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

… but whose IP core?

16

Page 17: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

• Mobile network operators (MNO) vs. virtual mobile network operators (MVNO)

• Infrastructure sharing between MNOs:

• Reduce deployment costs

• Roaming implementations: Home routed vs. Local Breakout

Layer 8

17

Page 18: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

A partial view of US cellular networks as seen by Netalyzr

VERIZON

AT&TSPRINT

T-MOBILE

BOOST

REPUBLICVIRGIN

GO SMARTFAMILY MOBILE

SIMPLE MOBILE

METROPCS

TRACFONE

WDSPCo

Hosted Data

Solutions

18

STRAIGHTTALK

Page 19: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Heterogeneous network properties!

Operator 1 (3G) Operator 2 (3G)

19

Invalid Host Field

DNS Proxy

Header/ContentModification

HTTP Enforcement

Percentage of sessions

0 25 50 75 100

Percentage of sessions

0 25 50 75 100

Home NetworkRoaming

Page 20: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Lessons learnt

• Measuring mobile performance is a difficult multidimensional analysis

• Network properties, and services are not homogeneous even within a single operator

• Middleboxes can affect performance and bias server-side performance measurements: measure from the edge for fidelity

• Layer 8 must be taken into account: roaming implementations and types of MVNO define the IP core

20

Page 21: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Future Work• Characterize mobile ecosystem (MNO and MVNO) at

a global-scale

• Get better insight on business relationships and network sharing agreements

• Characterization of IP core: location of DNS resolvers and proxies and their implications

• IPv6 - IPv4 performance

• Impact of mobile infrastructure on CDN performance and cloud services

21

Page 22: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Try Netalyzr and tell us what you’d like to know

about your network!

22

[email protected] !

[email protected] http://www.netalyzr.icsi.berkely.edu/publications

Page 23: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Mobile app designFront-end!(ICSI)

Back-ends!(Amazon EC2)

JSon - based report

Android activity

Background service

23

Page 24: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

App reviews and feedback

:) :(24

Page 25: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

MVNOs vs MNO

�����������

������

���

�����

����� ��

������� ��������

���� ��!���"�#��

$��

������� ��������

���� ��!���"�#��

$%��� &�� �&���'' �

���(��

���)����&����&*

&&����&*

��� �������

��������+�,������ �&�-

25

Page 26: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Netalyzr Facts (as of Sept. 2014)

• 1.4M total sessions executed in more than 170 countries

• Android version: 22K installs and 41K sessions executed

• Research outcome:

• Test-suite description and analysis of the state of the Internet [IMC’10]

• DNS interpositioning for error traffic monetization [USENIX FOCI’11]

• DNS status and performance [SATIN’11]

• UPnP vulnerabilities on home gateways and proxy deployments at a world-wide scale [PAM’12 and PAM’13]

• Analysis of Root certificates on Android handsets in the wild [CoNEXT’14]

26

Page 27: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

Roaming implementationsPDP CONTEXT

(APN SETTINGS)

VISITING NETWORK

UE

DATA-PLANE

DNSResolver

IP CORE

Proxy NAT

PDN

1. HOME ROUTED:

PDP CONTEXT(APN SETTINGS)

DATA-PLANE

PDN

VISITING NETWORK

UE

2. LOCAL BREAKOUT:

DNSResolver

IP CORE

Proxy NAT

CONTROL-PLANE

SGSNS-GW

CONTROL-PLANE

BILLING ANDAUTHENTICATION

GGSNP-GW

HOME NETWORK

RNCeNB

CONTROL-PLANE

SGSNS-GW

CONTROL-PLANE

BILLING ANDAUTHENTICATION

GGSNP-GW

HOME NETWORK

RNCeNB

27

Page 28: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

… and network deployment!3G 4G CDMA2000

100

60.339.70.499.6

0.299.8

1.598.5

100

100100

19.580.5

100

20.4 2.975.8 0.9

100100

100

100

4060595

88.911.1

100

2575

11.8 2.584.5 1.2

100

100100

100

100

10099 1

100100

44.655.4100

20 11.665.3 3.2

100

100

2575

100

67.432.6

100

100

964

100

100

100

100

59.140.9

OTHERSSWISSCOM (CH)

T−MOBILE (DE)VIRGIN (FR)

3 (UK)SPRINT (US)

SFR (FR)AT&T (US)

BOUYGUES (FR)ORANGE (FR)VERIZON (US)

T−MOBILE (US)FREE (FR)

OTHERSJOE (FR)

VIRGIN (FR)VIRGIN (US)BOOST (US)SPRINT (US)

SFR (FR)AT&T (US)

BOUYGUES (FR)ORANGE (FR)

METROPCS (US)T−MOBILE (US)

FREE (FR)

Hom

e Network

MVN

O/R

oaming

ED C D EDP ED C D EDP ED C D EDPTOPOLOGY

25 50 75 100Percentage of sessions per operatorand 3GPP standard

�������������

� � �����

��

����� ��������������� �����

�������������

� �

�����

��

��������������������� �����

�������������

� ������

��

���� �� ���� �����!��������� ������

�������������

�����"� �

��

#��� �� ���� �����!�� ��� ��������� �������

28

Page 29: Understanding user experience on mobile devices with the ...narseo/NANOG_Maryland.pdf• Native Android version since November 2013 (22K installs in 120 countries) • Research outcome:

The impact of network topology

0.00

0.25

0.50

0.75

1.00

0 500 1000 1500DNS lookup time for uncached names (ms)

ECD

F

Inappropriate network topology can double DNS lookup time!

29

3G

4G

Median DNS Lookup Time (ms)

0 150 300 450 600

DNS Before CGNDNS After CGN