Transcript Slide 1
Can We Pay for What We Get
in 3G Data Access?
Chunyi Peng, Guan-Hua Tu, Chi-Yu Li, Songwu Lu
University of California, Los Angeles
ACM MOBICOM 2012
Istanbul, Turkey
C PENG (UCLA) @ MOBICOM'12
Mobile Data Access is Popular
2
Core
Cellular
Network Internet
62% US broadband users with wireless data plans;
1.2 billion global users for mobile web.
C PENG (UCLA) @ MOBICOM'12
Mobile Data Accounting
3
Cellular Network
$$$
Internet
Usage-based charging based on
data volume
e.g., $15 for 200MB for AT&T iPhone
Accounting: How much data
is actually used?
C PENG (UCLA) @ MOBICOM'12
Accounting in 3G Networks
4
3G Cellular Network
RNC
UE
SGSN
GGSN
Internet
BS
Alice
VOP_RAW
Policy
VOP
• Accounting done at SGSN/GGSN
• Accounting policy defined by carriers
C PENG (UCLA) @ MOBICOM'12
2 Issues in 3G Accounting
5
RNC
UE
BS
SGSN
GGSN
Question: VUE = VOP?
VUE
Alice
VOP_RAW
1. VUE ≠
Policy
VOP_RAW?(accounting
architecture)
VOP
2. VOP_RAW ≠ VOP?
(policy practice)
Internet
C PENG (UCLA) @ MOBICOM'12
Contributions
6
First work to assess mobile data accounting
Largely successful, but pathological cases do exist
Study accounting discrepancy between the operator’s log
and the user’s record
Identify 2 extreme cases
WE PAY FOR WHAT WE DO NOT GET
WE GET WHAT WE DO NOT PAY FOR
Explore root causes
limitation in accounting architecture
Loopholes in policy practice
Suggest remedies
C PENG (UCLA) @ MOBICOM'12
Methodology
7
Conduct experiments over 2 US carriers
Partial
validation with 3rd US carrier and 2 operators
in China and Taiwan
Both extreme and common cases
Use Android phones for mobile data access in
various test scenarios
Accessing accounting records VOP from operators
#1:
Dial-in for the remaining monthly data usage
#2: Online itemized data usage
BillAudit: logging usage VUE @smartphones
C PENG (UCLA) @ MOBICOM'12
The Rest of Talk
8
“Overcharging”
Extreme
cases
Average cases
Root cause: limitation in 3G accounting architecture
“Undercharging”
Root
cause: Loopholes in policy enforcement
“Gray” areas
Discussion and summary
9
WE PAY FOR WHAT WE DO NOT GET
C PENG (UCLA) @ MOBICOM'12
Extreme Case: No Signal
10
✗
VUE
DL-NS experiment over UDP
VOP_RAW
3G Network
Server
Result:
(1) Issue a UDP-based service
s = 50Kbps, t = 10 mins
VOP ≈ VSR= 50K ✗
x10 x(2)
60/8
= to
3.75MB
Move
a blind zone
VUE ≈ 0
✗
UEs PAY FOR WHAT THEY
DO NOT
GET.
(3) UDP traffic
for t mins
(rate: s)
VUE
VOP
VSR
C PENG (UCLA) @ MOBICOM'12
How Bad the Gap Can Be?
11
Gap = VOP – VUE ≈ S × T
UDP source
S: 50Kbps ~ 8Mbps
Duration T: 1min ~ 6 hours
lasts at least three hours!
Observed gap reaches 450MB (t = 1h, s = 1Mbps)!
Operator-I, t = 1min
Source Rate (Mbps)
S = 50 Kbps
Time (hours)
C PENG (UCLA) @ MOBICOM'12
Root Cause
12
12
✗
RNC
SGSN
GGSN
✗
VUE ---
VOP
3G accounting decision takes local view at
SGSN/GGSN, w/o using feedback from end-host.
C PENG (UCLA) @ MOBICOM'12
Still-Bad Case: Even With Signals
13
DL-NS experiments with different signal strength
3G Network
RSSI (dBm)
-90
105
✗
Strong-Signal (SS-zone)
Weak-Signal (W-zone)
Server
(1) Issue a UDP-based service
(2) Move to a blind zone
(2) Stay in different zones
✗
(3) UDP traffic for t mins (rate: s)
Weaker-Signal (WR-zone) V
UE
-113
No-Signal (NS-zone)
VOP
VSR
C PENG (UCLA) @ MOBICOM'12
Gap Exists Even With Signals!
14
S , Gap
RSSI , Gap
Cause: Packet drops over radio link.
(Kbps)
Source Rate (Kbps)
UEs PAY FOR WHAT THEY DO NOT GET,
though wireless link exists!
C PENG (UCLA) @ MOBICOM'12
Still-Bad Case: Intermittent Signals
15
When users lose signals for a while but recover
them shortly
The gap exists with transient lost links
Buffering
and retransmission over radio links may
reduce the gap (see the paper)
UEs PAY FOR WHAT THEY DO NOT GET, when
they temporarily (10+ seconds) lose wireless links!
C PENG (UCLA) @ MOBICOM'12
So Terrible In Reality?
16
Good news: Probably not!
16
✗
✗ ✗
VUE ---
RNC
SGSN
GGSN
✗
VOP VOP--
✗
TCP/App control will teardown it (adjust its incoming rate)
Gap for DL-NS over TCP: 2.9 ~ 50KB
C PENG (UCLA) @ MOBICOM'12
Application Behaviors
17
DL-NS tests with 5 applications:
Web,
Skype, YouTube, PPS streaming, VLC streaming
over VPN
Web Skype YouTube PPS
Med (MB) -0.03 0.88
0.23
3.30
Min (MB) 0.00
0.40
0.20
0.72
Max (MB) -0.04 0.99
0.34
4.3
VLC
2.97
1.45
29.9
Mobile accounting is largely successful in practice.
Users may occasionally be overcharged
It depends on when and how app control works.
C PENG (UCLA) @ MOBICOM'12
Real User Performance
18
Two-week usage for 7 users
Operator-I
Operator-II
User
1
2
3
Apps.
Map
Stock
Game
VUE
194.2
VOP
Gap
4
5
6
7 (1day)
Skype, YouTu Ebook
PPS
be,
etc.
PPS
-
YouTub
e, PPS
270.3
124.6
900.2
121.7
47.1
72.4
192.6
270.0
129.4
948.4
120.9
47.3
77.6
-1.8
-0.3
4.8
48.2
-0.8
0.2
5.2
-0.9%
-0.1%
3.9%
5.3%
-0.6%
0.4%
7.2%
YouTube
on the train to NYC.
C PENG (UCLA) @ MOBICOM'12
3 Views on “Overcharging”
19
Optimistic view: not too bad in reality, no fix
Built-in TCP/application control is sufficient
Alternative (Operator’s) view: not to intend to
account the data volume to end-hosts, but the one
traversing the core network, no need to fix
Security: What if that the data is not what users want?
Audit: How to guarantee that inside accounting is correct?
Conservative view: need to fix it
Users should pay for what they get
3G accounting architecture should not depend on external
control
C PENG (UCLA) @ MOBICOM'12
Proposals
20
Exploit feedback from devices in accounting decision
E.g.,
using info already collected by cellular networks
VOPVOP - VRNC_unsent
VRNC_unsent
20
RNC
Packet drops
SGSN
GGSN
21
WE GET WHAT WE DO NOT PAY FOR
C PENG (UCLA) @ MOBICOM'12
22
Loopholes in Accounting Policy
Practice
RNC
SGSN
GGSN
BS
Loophole:
Policy + Loophole
• A DNS flow should be identified by
tuples
(src_addr,
dest_addr,
anyfive
fake
DNS
message,
or anyPolicy
dest_port,
protocol
ID)port
realsrc_port,
data packet
using
DNS
• But only dest_port (+ protocol ID) is
(53),
be free of charge!
usedcan
in practice
VOP (ANY-over-DNS) = 0
VOP_RAW
Policy:
Free DNS Service
VOP
VOP (DNS) = 0
C PENG (UCLA) @ MOBICOM'12
Our Findings
23
Free DNS policy enforcement
Operator-I:
Packets via port 53 are free
Operator-II: Packets via UDP+Port 53 are free
Exploit “DNS tunneling” for free data access
Proxy
server (outside 3G network) relays packets
to/from UE via Port-53
Observed: Free data access > 200MB, VOP = 0
No sign to limit “free” data volume
C PENG (UCLA) @ MOBICOM'12
More on Operator Policy
24
Other carriers
3rd US carrier: free DNS by June 2012, no free after July
China/Taiwan carriers: no free DNS service at all
Accounting policy is operator specific
Other free or differential-pricing policies
Free Internet access to a given website
Free access via a specific Access Point Name (APN)
Hack: web redirection for free Internet access
Hack: use this APN, not the default APN
Unlimited plans/discounts for Facebook access
Similar to web redirection if we can evade Facebook (probably not)
C PENG (UCLA) @ MOBICOM'12
Discussion and Proposals
25
Operators have freedom to define their own policy Policy
Flexibility to compete in the market
Gap between policy and policy enforcement
Should be conflict free
Otherwise, policy may open loopholes unanticipated
Simplest fix: stop free DNS service
Negligible DNS traffic volume in normal cases
Other options:
DNS server authentication
Quota
Message integrity check
26
“GRAY” ACCOUNTING AREAS
C PENG (UCLA) @ MOBICOM'12
Effect of Middle-boxes
27
Middle-boxes lead to inconsistent accounting
views at the core network and the end host
Pay
for the uplink to a non-existing host due to
FTP/HTTP proxy
RNC
SGSN
GGSN
Middl
e-box
Invalid
✗ link
✔V
OP
>0
✗
C PENG (UCLA) @ MOBICOM'12
Packet Drops over the Internet
28
Misbehaviors over the Internet can incur extra
mobile data charging
Packet
drops over the internet increases volume within
cellular networks
RNC
TCP ReTX
SGSN
GGSN
VOP
Packet drops
C PENG (UCLA) @ MOBICOM'12
Overhead for Wanted Content
29
VOP covers protocol overhead and app. signaling
HTTP redirection: #redirection , VOP
Email: significant protocol overhead for sending a short
email
Skype: significant protocol management overhead
VOP covers Ads, or whatever users may not expect
Hidden cost for free-version applications with more Ads?
Security issue?
Content-centric charging?
C PENG (UCLA) @ MOBICOM'12
Beyond Accounting
30
Revisit charging/accounting design principles
Cooperate
with Internet?
Segmented charging for one data service?
Who should pay?
Receiver-based, sender-based, or both (current
practice)?
For what?
Volume? Content? Part of content?
What if using different pricing schemes?
C PENG (UCLA) @ MOBICOM'12
Discussion and Future Work
31
Revisit accounting architecture
What
failures and losses should be handled?
What mechanisms are indispensable for given failures?
When and how does the end host report delivery
losses?
How to ensure that the feedback information is secure
and trustworthy?
How many mechanisms should be placed into the
future cellular network standards?
Policy and policy enforcement
C PENG (UCLA) @ MOBICOM'12
Summary
32
First assessment of mobile data accounting system
over operational 3G networks
Largely
successful, but also exceptions
Accounting discrepancy between the operator’s
log and the user’s record
Identify
two extreme cases:
WE
PAY FOR WHAT WE DO NOT GET
WE GET WHAT WE DO NOT PAY FOR
Explore
root cause in accounting architecture & policy
Propose remedy suggestions
Many research issues ahead
e.g.,
security, auditing, pricing, …