MOTA: Engineering an Operator Agnostic Mobile Service

qualtaghblurtingΚινητά – Ασύρματες Τεχνολογίες

12 Δεκ 2013 (πριν από 4 χρόνια και 19 μέρες)

76 εμφανίσεις

MOTA: Engineering an Operator Agnostic
Mobile Service

Supratim Deb, Kanthi Nagaraj, Vikram Srinivasan

Bell Labs

Mobile Data Explosion and Need for New Technological Innovations

FCC National Broadband Plan:



500 MHz of additional spectrum



Technical and Business innovations that increase efficiency of spectrum utilization

Wireless Service Today and the End User Perspective

Takeaways:



Spectrum shortage exacerbated by deployment practices



Users demand choice


Wireless service provider should depend on

location, pricing and user
preferences

Src: opensignalmaps.com

Challenges in User’s Making Appropriate Choices?


Option 1:

Centralized entity makes choices.


Operators unlikely to share network planning
information.



Option 2:

User’s use signal strength from
different base stations


This is insufficient and can result in poor user
experience.


Additional signaling information needed.


Choice of network should depend on user
mobility pattern


Switching at fine time scales incurs huge overhead
in core network



Goal:


Distributed decisions by each user


Concise network signaling that accounts for
mobility


Evolutionary over current standards.



Everyone
joins O2

VF may be
better choice

Src: opensignalmaps.com

MOTA Service Model


Service Aggregator: New intermediary
between users and operators


Responsible for maintaining customer
relationships


Handles all control plane operations that
cannot be handled by a single operator


Tracking and paging


Billing and authentication


Seamless switching across operators at Layer 3




CoreNet
-
1

PGW

BTS

BTS

CoreNet
-
2

PGW

BTS

BTS

Service Aggregator Cloud


AAA
Server

Tracking
& Paging

Mobile
IPv6
Anchor

Module
for
Switching
Decisions

Network layer
and above

MAC and lower
layers

MIH Layer

MOTA Framework


What information should each operator
maintain?





What aggregate information should be
broadcast by each base station?



What information should each user maintain?



How should a client decide the following:


What operator to associate with each interface
(2G, 3G, 4G)?


What applications to associate with each interface
(Voice, video, data etc.)?



User mobility?

Session duration?

User experience?

Network load?

Price?

Operator 1

Operator 2

Operator m

2G Interface

3G Interface

4G Interface

Application 1

Application 2

Application n

User behavior

Network experience

Battery status

Utilities and Proportional Fairness


The Framework for All Seasons!


User utility













User’s Objective:






Subject to:


Each application associates with only one interface


Each interface associates with only one operator.


a
a
a
a
a
a
a
p
R
w
p
R
U



ln
)
,
(
Weight of application
a

Rate of application
a

Price of application
a

Price sensitivity of application
a

)]
,
(
[
a
a
A
a
a
p
R
U
E
Maximize


Comment: Price for each operator is constant. Operators sets a single price per unit

weight per technology across all cells

Signaling and Algorithm for Static Clients


Fact:



Proportional fair scheduling is typically used by most cellular technologies.



If total weight of applications associated with a base station

j

is
W
j
, and PHY rate of user
u

is
r
u

and weight of his application is
w
a
,
then aggregate rate user receives under proportional fair
scheduling is:











Network Signaling for Static Users:
Each base station only needs to transmit its
aggregate load
W
j

and its price
p
j.


r
j
a
j
r
W
w
T

Recall



Operator 1

Operator 2

Operator m

2G Interface

3G Interface

4G Interface

Application 1

Application 2

Application n

Base station conveys

1.
Price per unit weight

2.
Total load

User computes

1.
Which operator to select for each technology

2.
Which application goes to each technology


Based on

1.
Signaling information

2.
Energy considerations

3.
Application characteristics

Greedy User Algorithm


Utility of associating application of weight
w

to base station
j
=
w

f(p
j
, W
j
, w)


Utility of operator that offers maximum utility is

G
l





Order application weights in increasing order

w
1

<= w
2
,… <= w
n


Assign applications in this order.


Greedy Algorithm:


Iterate over all applications


In the rth step


Assign application r to interface that maximizes







)
,
,
(
max
)
(
w
W
p
f
w
G
j
j
operators
l

)
_
(
)
_
(
)
_
(
)
_
(
weight
curr
G
weight
curr
w
weight
curr
G
w
weight
curr
r
r



Price of Anarchy


Global Efficiency versus Selfish Strategy


Theorem:


Let
r
be vector of PHY data rates of all users.


There exists a constant K, such that












Comment:
Proportional fair scheduling at base stations ensures that local
decisions are not very bad.




K
1
)
(
)
.
(
r
GLOBAL
r
K
SELFISH

Signaling for Mobile Users


Question:

What signaling information should the base station send that is useful from a
user’s perspective?



Answer:

Something that will allow the user to compute her net utility when she
associates with this operator and moves around.



Question:

Isn’t this dependent on each user’s individual mobility pattern?



Answer:
Clearly yes. Hence convey only aggregate information based on average usage
patterns. This could depend on time of day etc.

Signaling for Mobile Clients


Base station tracks:





= aggregate
log(PHY rate)

over the time spent in
cell
-
k

by user u’s
application
a
,

when it is initiated in
cell
j
.







= aggregate time spent in cell k,
by users u’s application

a
initiated in cell
-
j




For each application class, Base station
k

conveys:



)
(
,
a
R
k
j
Cell j

Cell k

)
(
,
a
T
k
j


k
k
j
k
k
j
a
T
E
W
a
R
E
)]
(
[
)
ln(
)]
(
[
,
,
User Utility and Algorithm


Recall user utility in static case =
w

f(p
j
, W
j
, w)



In mobile case =

/(application duration)




Assumption: Total load at base station much larger than individual weight of user
applications



Can now apply standard Maximized Generalized Assignment algorithm



E.g.: Local Greedy Search with ½
-

e

factor approximation.



k
k
j
k
k
j
a
T
E
W
a
R
E
)]
(
[
)
ln(
)]
(
[
,
,
Static algo cannot be
applied

Difficult to quantify price of anarchy. In mobile case, scenario is more dynamic.

Similar to multiple agent learning. Difficult to prove strong guarantees.

Putting it together in practice

Implementation over Existing IEEE, IRTF and IETF proposals:


Use IEEE 802.21 for signaling



IRTF MPA framework for authentication and acquiring IP address and network
resources.


Fast Handover in MIPv6 to simultaneously establish tunnel to gateway of new network
and forward packets.


Gathering network state information:


Needs to be managed carefully depending on FDD versus TDD systems to minimize
overhead.

Evaluation

Network Topology:


Cell tower location of a major operator in
Indian city (5Km X 5Km area)


Clutter information along with RF tool used to
generate RF map


We assume two operators share the same cell
tower locations.


Each offers HSDPA and LTE

Application Models:


3 classes, voice, video and data


Generated according to guidelines for next
generation mobile networks


User Mobility:


Manhattan and random waypoint

Performance Improvement as Fraction of Mobile Users is Varied



Area Spectral Efficiency improves by 2.5X
-
4X

Performance Gain over Optimized Single Operator



At least 60% gain over single operator with load balancing across technologies

What’s in it for the Operators?

Price

User Utility

Traditional Model

MOTA Model

Operator incentive

Simulations imply 20% incentive. Far more research required.

Reflections


Are there alternative simpler
architectures possible that just
exploit roaming agreements between
operators?



How can this be combined with ideas
of dynamic spectrum access? Do
operators really need to swap
spectrum at fine time scales?



Is operator signaling really required?
Can end users learn appropriate
association over time?


A phone app that makes these decisions
for you.

Questions?