Slides - Euroview 2012

FUTURE MOBILE COMMUNICATION:
FUTURE
 
MOBILE
?
COMMUNICATION:
?
LTE OPTIMIZATION AND MOBILE NETWORK 
VIRTUALIZATION
VIRTUALIZATION
Yasir Zaki
,
 Andreas Timm‐Giel
,
 Carmelita Gör
g
,
,g
ComNets
University of Bremen, Technical University of Hamburg
EuroviewJuly 23
rd
2012
ComNets
yzaki@tzi.de
Outline

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
2

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
3
Wireless Virtualization

Wired virtualization is well‐known

A natural extension from wired to 
wireless virtualization

Virtualization of the wireless 
air interface is a scheduling problem:
air
?
interface
?
is
?
a
?
scheduling
?
problem:
 Tx/Rx power
 Frequency, Time, Code 
S ll ti

S
pace a
ll
oca
ti
on

Similar to the well known wireless 
transmission strategies: FDMA, TDMA, 
CDMA and SDMA
CDMA
?
and
?
SDMA
?

But virtualization is doing more
ComNets
yzaki@tzi.de
4
Motivation Behind Mobile Network Virtualization

Mobile networks are one of the fastest growing technologies
 has big influence on our daily activities

Often, the wireless resources of mobile networks are expensive and scarce

䉩 扬 t h d 瑩 i 瑨 i 桩 桬 瑩 瑩

B
e
i
湧  a
bl
e 
t
o s
h
are an
d
 op
ti
m
i
ze 
th
e resources usage 
i
s 
hi
g
hl
y mo
ti
va
ti
ng

Network virtualization is a good solution :
 reduces the number of base stations (reduce energy usage)
 allows completely new value chains  (mainly  smaller players)

In addition, sharing the frequency resources among multiple operators is 
very appealing
 gives operators the flexibility to expand/shrink their networks on the fly
ComNets
yzaki@tzi.de
5
LTE Virtualization

Virtualizing the LTE network means to virtualize the infrastructure of 
the LTE system

This allows multiple network operators to create their own virtual 
network depending on their requirements and goals

The challenges are:
 How to virtualize the 
p
h
y
sical infrastructure?
p y
 What kind of changes are required to the existing LTE system?

坥 浡楮汹 景牥獥f 瑷t 摩晦敲敮d 瑹灥t 潦 癩牴畡汩穡瑩潮 灲潣敳獥猺

坥
 
mainly
?
foresee
?
two
?
different
?
types
?
of
?
virtualization
?
processes:
 Virtualizing LTE physical nodes (e.g. eNodeBs, routers, Ethernet links)
 Virtualizing the air interface of the LTE system (focus of this presentation)
ComNets
yzaki@tzi.de
6

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
7
LTE Air Interface Virtualization

In order to virtualize the LTE air interface, the eNodeB (LTE base station) has 
to be virtualized

Our solution is inspired by the XEN
1
architecture, where mainly a 
“Hypervisor” is added on top of the physical resources.

The hypervisor is responsible for virtualizing the eNodeB and scheduling the 
air interface (OFDMA) resources among the Virtual Operators (VOs)
ComNets
yzaki@tzi.de
8
1
XEN is a node virtualization software
LTE Hypervisor

It is responsible for:
 virtualizing the eNodeB

scheduling the air interface resources among the
Physical
Resources
PRBs
Physical eNB
scheduling
 
the
?
air
?
interface
?
resources
?
among
?
the
?
Virtual Operators (VOs)

䥴 捯汬散瑳 慬a 牥汥癡湴 楮景牭慴楯i ⡦牯( 慬a
Hypervisor (2nd level Scheduler)
Virtual eNBs
Channel
Conditions

It
 
collects
?
all
?
relevant
?
information
?
(from
?
all
?
VOs) :
 users channel conditions
 traffic load
Virtual

eNBs
LTE MAC
Schedule
r
LTE MAC
Schedule
r
LTE MAC
Schedule
r
 VO requirements
 VO contracts
 etc.

Based on that, the hypervisor tries to allocate 
the resources to the VO
ComNets
yzaki@tzi.de
9

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
10
OPNET Simulation Model
H
yp
erviso
r
Physical eNB
yp
Virtual
eNBs
ComNets
yzaki@tzi.de
11

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
12
LTE Contract‐based Spectrum Management

We defined four different types of contracts that the infrastructure 
provider offers to the virtual operators and these are:
a)
Fixed guarantees
a)
Fixed
 
guarantees
b) Dynamic guarantees
c) Best effort (BE) with min guarantees
d)
Best effort with no guarantees
d)
Best
 
effort
?
with
?
no
?
guarantees
ComNets
yzaki@tzi.de
13
Continue … 

In order for the hypervisor to be able to satisfy the operator requests and 
their predefined contracts, an estimate of the actual needed spectrum of 
each operator is required

The operators need to feedback this estimate value back to the hypervisor 
(in a predefined time interval)

The PRBs estimate of each operator can be calculated iteratively as follows:
   
)1(1)(




 nPRBsTTInEnE
Where:
 E(N) is the averaged required PRBs count (estimate of the required bandwidth) over N time 
interval
 PRBsTTI(N) is the instantaneous PRBs count at the Nth TTI calculated by summing the PRBs 
that were additionally needed to schedule the un‐served users within this TTI
 N is the number of TTIs in the hypervisor interval

α is the smoothing factor indicating the weighting
ComNets
yzaki@tzi.de
14
α
 
is
?
the
?
smoothing
?
factor
?
indicating
?
the
?
weighting

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
15
Simulation Configurations

The simulation is configured with 4 virtual operators each is 
configured with one of the different contract types defined earlier:
Virtual Operator Contract details
VO1 (Video streaming) “fixed guaranteed contract” of 33 PRBs
V
O2 (VOIP) “dynamic guaranteed contract”, with a max value of 33
PRBs
VO3 (VOIP + BE Video on demand) “best effort with min. guarantees” contract, with min. and
max. value of 25 and 45 consecutively

Two scenarios are configured one without virtualization “legacy” and 
VO4 (Small VOIP operator) “BE and no guarantees” contract
one with virtualization “virtualized”.
ComNets
yzaki@tzi.de
16
Simulation Configurations
Parameter Assumption
Number of VO 4
Number of virtual
eNodeBs
4 eNodeBs
(one per VO)
Number

of

virtual
eNodeBs
4

eNodeBs
(one

per

VO)
Total number of PRBs 99 (corresponds to about ~ 20 MHz)
Number of active users VO1: 12 video users
VO2: 40 VOIP users
VO3: 16 VOIP + 16 video users
VO4: 3 VOIP users
Mobility model Random Way Point (RWP) with 5 km/h
Channel
model
ITU
Ped
-
A
Channel
model
ITU
Ped
A
DL VOIP traffic model Silence/Talk Spurt length = neg. exp. with 3 sec mean. Call duration = uniform (1, 3 min)
Inter-repetition time = negative exponential with 90 sec mean
DL Video traffic model Video conferencing application with unlimited duration
In/Outgoing streaminter arrival time = Const (0 01 sec)
In/Outgoing

stream

inter

arrival

time

=

Const

(0
.
01

sec)
In/Outgoing stream frame size = Const (80 Bytes)
Hypervisor resolution,
estimation factor α
1 sec, with
0.5
Simulation time 1000 sec
ComNets
yzaki@tzi.de
17
Virtual Operator (VO) allocated number of PRBs

The figure shows the number of 
PRBs that each VO has been 
allocated over time

It can be noticed that for the first 
operator the PRBs allocation is fixed
operator
?
the
?
PRBs
?
allocation
?
is
?
fixed
?
to 33 PRBs since it is using the fixed 
guaranteed contract

For the other three operators we 
can notice that the allocated 
number of PRBs changes with time
number
?
of
?
PRBs
?
changes
?
with
?
time
?
depending on the traffic load and 
the contract details of each 
o
p
erator
ComNets
yzaki@tzi.de
18
p
Virtual operator 1 (12 video users)
Air interface throughput and app. end‐to‐end delay
What
can
be
noticed
is
that
the
operator
has
the
same
performance
with
and
without
What
can
be
noticed
is
that
the
operator
has
the
same
performance
with
and
without
virtualization;this is because this operator has a contract with a guaranteed fixed
allocation
ComNets
yzaki@tzi.de
19
Virtual operator 2 (40 VOIP users) 
Air interface throughput and app. end‐to‐end delay
What
can
be
noticed
is
that
the
operator
has
the
same
performance
with
and
without
What
can
be
noticed
is
that
the
operator
has
the
same
performance
with
and
without
virtualization
ComNets
yzaki@tzi.de
20
Virtual operator 2
Downlink used number of PRBs vs. time
The previous results showed that operator 2 has the same performance with and without
virtualization
But,in the “virtualized” case VO2 is not wasting the air interface resources since it only uses
the required number of PRBs to serve the users as can be seen above
This is a big advantage since the operator will be able to cut cost because he will only pay for
ComNets
yzaki@tzi.de
21
the resources used
Virtual operator 3 Application end‐to‐end delay
We can see that similar performance is achieved in both cases for the VOIP users (left side).
As
for
the
video
users
(right
side),
they
are
suffering
from
huge
delay
in
the
“
legacy
”
scenario
As
for
the
video
users
(right
side),
they
are
suffering
from
huge
delay
in
the
legacy
scenario
due to buffering;whereas in the “virtualized” scenario they are having good performance.
The reason why the VOIP users in the “legacy” scenario are not affected is the fact that these
users
are
being
served
with
higher
priority
and
the
resources
are
enough
to
serve
those
users,
ComNets
yzaki@tzi.de
22
users
are
being
served
with
higher
priority
and
the
resources
are
enough
to
serve
those
users,
but not enough to serve the video users.
Virtual operator 4 (3 VOIP users)
Air interface throughput and app. end‐to‐end delay
One additional advantage that can be achieved in the “virtualized” scenario is the ability to
serve
small
operators
with
relatively
smaller
number
of
users
in
a
pure
best
effort
manner
with
serve
small
operators
with
relatively
smaller
number
of
users
in
a
pure
best
effort
manner
with
whatever resources are left rather than wasting these resources
ComNets
yzaki@tzi.de
23

Motivation of Wireless and LTE Virtualization

䱔L 噩牴畡汩穡瑩潮 䙲慭敷潲F

䱔L
 
Virtualization
?
Framework

Simulation Model

Contract Based Framework

Simulation Configurations and Results

Conclusion and Outlook
ComNets
yzaki@tzi.de
24
Conclusion and Outlook

The simulation results showed that a better performance can be achieved by 
using network virtualization in the LTE system.

Based on the contract configurations and the traffic load of each virtual 
operator the air interface resources are shared among the VOs.

The overall resource utilization is enhanced and the performance for both 
the network and the end‐user is better. 

Both operator 2 and 3 benefited from virtualization mainly by being able to 
cut costs and providing better performance for their users. 

The results also showed the possibility of opening the market to new players 
(small operators) that can serve a specific role and have small numbers of 
ComNets
yzaki@tzi.de
25
users.
THANKS FOR LISTENING
THANKS
 
FOR
?
LISTENING
ANY QUESTIONS
ComNets
yzaki@tzi.de
BACKUP SLIDES
ComNets
yzaki@tzi.de
Long Term Evolution (LTE)

LTE is the latest evolution of 3GPP standard. 

Its based on Orthogonal Frequency Division

Its
?
based
?
on
?
Orthogonal
?
Frequency
?
Division
?
Multiple Access (OFDMA) in the downlink 
and Single Carrier FDMA (SC‐FDMA) in the 
uplink
uplink

Its a very good candidate to be considered 
for applying network virtualization

䱔L
’
s湥n 湥瑷潲n 慲捨楴散瑵牥 楳 扡獥b 潮

䱔L s
 
new
?
network
?
architecture
?
is
?
based
?
on
?
a cost efficient two nodes architecture
 Enhanced NodeB (eNodeB) 

Access Gateway (AGW)
ComNets
yzaki@tzi.de
28

Access
 
Gateway
?
(AGW)
BE Virtual Operators Allocation

The allocation of the left PRBs to the BE operators in step 4 will 
be done based on a Fair Factor (FF) which is defined as follows:
 




VOBE
i
i
E
nE
FF
_#


f
ll
i



j
j
n
E
1
Where:


P
RBsLe
f
t
F
Fa
ll
oc
P
RBs
ii
_
i
nt
_


# BE_VO is the number of best effort virtual operator
 PRBs_alloci is the allocated PRBs for operator i
 Left_PRBS is the number of PRBs left after allocating VO of contract type a, b and 
h f
ComNets
yzaki@tzi.de
29
t
h
e min guarantees o
f
 type c
Virtual Radio*

Defines a framework for configurable 
radio networks

I d h k 楬i i

I
t exten
d
s t
h
e networ
k
 v
i
rtua
li
zat
i
on 
concept into the wireless domain known 
as “radio virtualization” 

Different virtual radio networks can

Different
?
virtual
?
radio
?
networks
?
can
?
operate on top of a common shared 
infrastructure and share the same radio 
resources

It presents how the radio resource sharing 
can be performed efficiently without 
i t f b t th diff t i t l
i
n
t
er
f
erence 
b
e
t
ween 
th
e 
diff
eren
t
 v
i
r
t
ua
l
 
radio networks
*
J Sachs S Baucke
“
Virtual Radio
-
AFramework for Configurable
ComNets
yzaki@tzi.de
30

J
.
Sachs
,
S
.
Baucke
,
Virtual

Radio
A

Framework

for

Configurable

Radio Networks”; WICON’08, Hawaii, USA, Nov. 2008.
VANU MultiRAN*

V
䵬tiRAN™ 噩 t lB

V
慮a ‐
M
u
ltiRAN™
 
Vi
r
t
ua
l
 
B
ase 
Station is a commercial 
software 

Taking advantage of Vanu 
software RAN technology, 
M ltiRAN d l d t
M
u
ltiRAN
 was 
d
eve
l
ope
d
 
t
o 
support multiple virtual base 
stations (vBTS) running on a 
single BTS hardware platform
single
?
BTS
?
hardware
?
platform
. 
The expense of antennas, BTS 
electronics, and backhaul can 
all be shared.
all
?
be
?
shared.
* J. Chapin; “Overview of Vanu Software Radio”;
ComNets
yzaki@tzi.de
31
from http://www.vanu.com, June. 2009.
LTE Downlink Physical Resource Structure
Inter-carrier subspacing
15 kHZ
Sub-carrier
Frequency (Hz)
12 Subcarriers
12*15k=180kHz
ComNets
yzaki@tzi.de
32