Embed Size (px)
Citation preview
Mobile Networks and Applications (MONET) manuscript No.(will be inserted by the editor)
LTE Mobile Network Virtualization
Exploiting Multiplexing and Multi-User Diversity Gain
Yasir Zaki · Liang ZhaoCarmelita Goerg · Andreas Timm-Giel
Received: date / Accepted: date
Abstract Network virtualization is receiving immense attention in the re-search community all over the world. There is no doubt that it will play asignificant role in shaping the way we do networking in the future. There havebeen different approaches to virtualize different aspects of the network: someare focusing on resource virtualization like node, server and router virtual-ization; while others are focusing on building a framework to set up virtualnetworks on the fly based on different virtual resources. Nevertheless, one veryimportant piece of the puzzle is still missing that is ”Wireless Virtualization”.The virtualization of the wireless medium has not yet received the appropriateattention it is entitled to, and there have only been some early attempts in thisfield. In this paper a general framework for virtualizing the wireless mediumis proposed and investigated. This framework focuses on virtualizing mobilecommunication systems so that multiple operators can share the same physicalresources while being able to stay isolated from each other. We mainly focus onthe Long Term Evolution (LTE) but the framework can also be generalized tofit any other wireless system. The goal of the paper is to exploit the advantagesthat can be obtained from virtualizing the LTE system, more specifically vir-tualizing the air interface (i.e. spectrum sharing). Two different possible gainareas are explored: spectrum multiplexing and multi-user diversity.
Keywords LTE · Wireless Network Virtualization · Future Internet
Yasir Zaki · Liang Zhao · Carmelita GoergCommunication Networks, TZI, University of Bremen, Otto-Hahn-Allee, NW128359 Bremen, GermanyE-mail: yzaki,zhaol,[email protected]
Andreas Timm-GielInstitute of Communication Networks, Hamburg University of Technology21073 Hamburg, GermanyE-mail: [email protected]
2 Yasir Zaki et al.
1 Introduction
Virtualization is a well known technique that has been used for years, especiallyin computer science like the use of virtual memory and virtual operating sys-tems. What is new though is the idea of using virtualization to create completevirtual networks. One option in the Future Internet is the possibility of havingmultiple co-existing architectures, in which each is designed and customized tofit one type of network with specific requirements. Network Virtualization willplay a vital role in diversifying the Future Internet into, e.g., separate virtualnetworks that are isolated from each other and can run different architectureswithin. This paper is organized as follows: section 2 gives a short introductionto network virtualization and more specifically wireless virtualization and themotivation behind it. Section 3 introduces the Long Term Evolution, and ourproposal for mobile system virtualization is discussed. A network simulatorthat is developed in OPNET for the mobile system virtualization is describedin section 4. Section 5 shows the simulation scenarios and results, and finallyin section 6 the conclusion is discussed.
2 Network Virtualization
Due to the inherent flaws, e.g., inadequate support of security, mobility andmulti-homing, the current Internet architecture is continuously being extendedand many people believe that the current Internet will not be able to cope withall the requirements coming in the future. In order to satisfy the multitude ofnew services running on the network, many research activities on the FutureInternet architecture have been launched throughout the world, for example4WARD [1] in Europe, VINI [2] and GENI [3] in the U.S. and AKARI [4] andAsiaFI [5] in Asia. VINI is a virtual network infrastructure based on Planet-Lab on which researchers can deploy, run and test their own protocols andservices in a large scale. GENI (Global Environment for Network Innovations)is a novel suite of architectures which support a range of experimental pro-tocols, with virtualization as one of its most important features. The AKARIArchitecture Design Project aims to implement the basic technology of a newgeneration network from a clean slate, and network virtualization is seen asone of the principles. Asia Future Internet Forum (AsiaFI) was founded tocoordinate research and development on Future Internet and also here net-work virtualization is one of the main research topics. By observing theseprojects, one tendency can be perceived that network virtualization is an at-tractive technique which receives more and more research attention, and itwill be a key area in the future network development. In principle, networkvirtualization enables multiple virtual networks to coexist on a common in-frastructure. Each virtual network is operating similar to a normal networkand does not necessarily have the awareness of the underlying virtualizationprocess. Individual virtual networks can contain operator-specific protocolsand architectures which can be totally different from other co-existing virtual
LTE Mobile Network Virtualization 3
networks. Additionally, the virtualized networks can be used as componentsof today’s networks.
2.1 Motivation of Wireless Network Virtualization
Mobile system virtualization can have several advantages:
– For infrastructure providers: Large companies have to play both roles: sys-tem operator as well as infrastructure provider. With virtualization, theinfrastructure providers can concentrate on the maintenance of the physi-cal equipment and save manpower for running the networks.
– For virtual mobile operators: deploying and operating a mobile system suchas GSM or UMTS require enormous investments. Infrastructure sharing isvery attractive, where the huge investment on the hardware and funda-mental construction could be saved. It also enables the small companies toenter the market without huge investments. The deployment, maintenance,migration and upgrade of virtual mobile systems will be flexible and evenon-the-fly.
– For end users: the increased number of operators will bring a diversity ofservices to end users. This mean more options to satisfy the user’s personaldemand and subsequently enjoy the services with reasonable pricing.
2.2 State-of-the-art of Mobile System Sharing/Virtualization
One of the interesting commercial products of soft radio is VANU [6], that is awireless infrastructure solution that enables individual base stations to simul-taneously operate GSM, CDMA, iDEN and beyond. It is also announced thatthe virtualized base station and RAN are also supported by their products.Spectrum sharing is one crucial part of our proposal on the mobile system vir-tualization, through which multiple operators can be introduced into one band.Other than the traditional DSA (dynamic spectrum allocation) model, [7] pro-posed a centralized spectrum broker that has the responsibility to coordinatethe frequency allocation among different wireless networks. Two optimizationproblems, maximized requirements and minimized interference, are formulatedand algorithms are designed to solve the efficiency problem. One thing has tobe mentioned here that MVNO (Mobile Virtual Network Operator) has beenavailable for long time in the current mobile systems. However, from the def-inition of MVNO we know that they don’t own any physical resources andhave no impact on the network configuration or algorithms. Actually, MVNOsact more like a service provider. Our proposal introduced in the next sectionis trying to decouple the physical system from the network, and the VirtualOperator (VO) has the overall control of its own virtual network. The mobilesystem infrastructure sharing also exists along with the deployment of the net-work, especially at the beginning. The reason is to reduce the cost and to rollout the infrastructure quickly. As referred to in [8], the infrastructure sharing
4 Yasir Zaki et al.
has mainly three areas: Passive component sharing, Active element sharingand Geographical sharing. To reduce the operating cost, different operatorswill share the fundamental establishments like roof locations, tower frame,equipment houses and power supply. This kind of sharing doesn’t involve net-work virtualization at all and is already accepted by most of the operatorsbecause of the economical benefits. Geographical sharing is simply dividingthe whole area into several regions and each operator will be in charge of oneof them. In this way the full coverage can be achieved in short time by thefederation of operators on different regions.
3 LTE Virtualization Framework
Long Term Evolution (LTE) has been seen as one of the major solutions ofnext generation mobile systems. According to the specifications from 3GPP,LTE will provide much higher data rates compared to UMTS, low latencyand enhanced QoS especially for the users at the cell edge. Due to thosepromising system performances, we choose the LTE system as a case studyto show the advantages of (wireless) network virtualization. The basic ideais to virtualize the LTE base station or what is also known as the enhancedNode-B (eNodeB). The eNodeB is the physical hardware that is responsiblefor sending and receiving data to and from the LTE users. Virtualizing theeNodeB is similar to any other node virtualization, where an entity called”Hypervisor” is added on top of the physical resources, and is responsible forscheduling these resources between the different virtual instances running ontop. Fig 1 shows the LTE eNodeB virtualization architecture.
Physical Hardware
Hypervisor
Virtual eNodeB
PDCP
RLC
PHY
GTP-U
UDP
100 BaseT
IP
EthernetMAC
Virtual eNodeB
PDCP
RLC
PHY
GTP-U
UDP
100 BaseT
IP
EthernetMAC
Virtual eNodeB
PDCP
RLC
PHY
GTP-U
UDP
100 BaseT
IP
EthernetMAC
Sp
ectr
um
Co
nfi
g.
& B
W e
stim
atio
n
Mux/DeMux
Sp
ectr
um
Co
nfi
g.
& B
W e
stim
atio
n
Sp
ectr
um
Co
nfi
g.
& B
W e
stim
atio
n
Spectrum allocation Unit
Fig. 1 Virtualized LTE eNodeB protocol stack
The hypervisor is also responsible for scheduling the air interface resourcesor the LTE spectrum between the different virtual eNodeBs (different vir-tual operators). The virtual operators share the spectrum based on differentcriteria. The Hypervisor collects information from the individual virtual eN-odeB like: user channel conditions, loads, priorities, QoS requirements and
LTE Mobile Network Virtualization 5
information related to the contract of each of the virtual operators. This in-formation is used to schedule the air interface resources between the differentvirtual operators. LTE uses OFDMA in the downlink, this means that thefrequency band is divided into a number of sub-bands each with a carrier fre-quency. The air interface resources that the hypervisor schedules are actuallythe Physical Radio Resource Blocks (PRB). A PRB is the smallest unit thatthe LTE MAC scheduler can allocate to a user. Scheduling the PRBs betweenthe different virtual eNodeBs actually means splitting the frequency spectrumbetween the different eNodeBs of the different operators. The hypervisor canmake use of apriori knowledge (e.g. users channel conditions, virtual operatorcontract, load ... etc.) to schedule the PRBs. OFDMA scheduling has beenstudied extensively in the literature [9] [10] [11] [12], but what is new here isthe scheduling of the frequency spectrum among the different operators. Thisis even more challenging because of the additional degree of freedom that hasbeen added. A number of possibilities exist here where the scheduling could bebased upon different criteria’s: bandwidth, data rates, power, interference, pre-defined contracts, channel conditions, traffic load or a combination of them.At the end, the hypervisor has to convert these criteria into a number of PRBsto be scheduled to each operator, but the challenge is to make sure that theallocated PRBs would be fair and enable the operators to satisfy their require-ments.
Some mechanisms and guidelines has to be defined to guarantee the re-sources sharing between the operators for example setting a guaranteed amountfor each operator and leaving the rest of the resources to be shared. In addi-tion, choosing the correct time frame that the hypervisor operates with is animportant issue in order to guarantees the pre-defined requirements. In ourpaper, we only concentrate on two different types of the scheduler algorithms,a static and a dynamic one. In the static algorithm the spectrum is divided be-tween the virtual operators beforehand, and each operator gets his operatingspectrum and keeps it for the whole time. This is similar to today’s network,where each operator has his own frequency spectrum and no other operatoris allowed to use it. In the dynamic algorithm, the resources are allocated tothe different operators during runtime, and the amount and allocation canchange over time depending on the operators traffic load. The latter algorithmis an example of what can be gained by applying network virtualization intowireless mobile communication systems, where not only operators share thephysical infrastructure but will also share the frequency spectrum and this inturn will lead to better resource utilization.
4 Simulation Model
The LTE simulation model used in this paper was developed using the OPNETsimulation tool [13]. The model is designed and implemented following the3GPP specifications.
6 Yasir Zaki et al.
Fig. 2 Four Virtual Operators example of a virtualized LTE network
The focus of this work was not on the node or link virtualization, but ratheron the air interface virtualization and how to schedule the air interface re-sources among the different virtual operators, no node/link virtualization wassimulated, instead the assumption is that we have a perfect node/link virtual-ization. Figure 2 shows an example scenario, it can be seen that an additionalentity between the virtual eNodeBs - that is the “Hypervisor” - is added. Thisentity is responsible for scheduling the air interface resources (frequency spec-trum or PRBs) among the different virtual eNodeBs. The Hypervisor also hasdirect access to the MAC layers of each of the LTE virtual eNodeBs to collectthe required relevant information to be used for the scheduling like: operatorusers channel conditions and operator traffic load.
In our implementation, different versions of the hypervisor exist based onthe scenario type. In the multiplexing used case two version are implemented:
1. Static version: where the hypervisor allocates the PRBs among the differentvirtual operators just once at the beginning of the simulations. The numberof the allocated PRBs for each operator is equal, where each virtual eNodeBwill get the exact same amount of PRBs and keeps it regardless if it is beingactually used or not.
2. Dynamic version: where the PRBs are allocated to the different virtualoperators in a dynamic manner at equal time intervals. The amount of theallocated PRBs will depend on the load that each operator is experiencingduring the last time instance. In this way, each operator will only get hisrequired share of the PRBs and less waste of resources will occur.
In the Multi-user diversity use case, two different versions of the hypervisorare also implemented as follows:
LTE Mobile Network Virtualization 7
1. Static version: where each virtual operator is configured with a fixed amountof spectrum throughout the whole simulation time. The spectrum alloca-tion is localized and does not change.
2. Dynamic version: each operator is still configured with a fixed amountof spectrum throughout the whole simulation time similar to the staticversion. The difference here is that the spectrum allocation is no morelocalized but dynamically changing, i.e. different PRBs are allocated toeach virtual operator at different points in time depending on the operatorusers channel conditions, but the total amount of the allocated PRBs toeach operators is same as in the static scenario.
5 Simulation Configurations and Results
In this paper, we mainly aim to show the effects and advantages of usingnetwork virtualization in mobile communication systems. The goal is to showdifferent aspects and sides of network virtualization where some potential gainexists. Specifically we investigate what additional benefits can arise if the mo-bile operators actually share the air interface resources (i.e. frequency band).From that, the simulations scenarios investigated are divided into two differentuse cases: Multiplexing use case and Multi-user diversity use case. In the nextsubsections each of the use case are discussed.
5.1 Multiplexing Use Case
This use case investigates the gain that can be achieved from virtualizing theLTE air interface through multiplexing i.e. sharing the spectrum. The motiva-tion behind such a use case comes from the fact that within today’s networkeach mobile network operator is not fully utilizing the whole spectrum byhimself all the time. Nevertheless, the spectrum allocation is already prefixed,where each operator has already bought a fixed amount of spectrum. Thismeans that in some cases the air interface resources are not fully used andtherefore wasted. The use case shows that by sharing the spectrum betweenmultiple mobile network operators the free resources can be used by otheroperators leading to a better spectrum utilization. To demonstrate that, twodifferent scenarios are investigated within this use case as follows:
– Static hypervisor scenario: which could be viewed similar to today’s mobilenetwork setup apart from sharing the infrastructure which is an additionalbenefit
– Dynamic hypervisor (load based) scenario: which aims to show how themobile network operators can share the spectrum and what are the benefitsof the LTE virtualization
Now besides of the hypervisor scheduler algorithm used in each scenariothe rest of the configuration is exactly the same for both scenarios and is shownin table 1.
8 Yasir Zaki et al.
Table 1 Simulation configurations
Parameter AssumptionNumber of virtual operators 3 virtual operators each with one eNBeNodeB coverage area Circular with radius = 375 meters
3 sectors, each with 120 degreeTotal Number of PRBs 99 (which corresponds to about 20 MHz)Mobility model Random Way Point (RWP)Users speed 5 km/h
Number of active users
Cell1 Cell2 Cell3VO1 10 VoIP 10 VoIP 10 VoIP
4 Video 2 Video 1 VideoVO2 10 VoIP 10 VoIP 10 VoIP
4 Video 2 Video 1 VideoVO3 10 VoIP 10 VoIP 10 VoIP
4 Video 2 Video 1 Video
Path loss model 128.1 + 37.6 log10(R) dB, R in km [14]Slow Fading model Log normal distributed
Mean value = zeroStandard deviation = 8 dBCorrelation distance = 50 meters
Fast Fading model Jake’s modelCQI reporting IdealLink-to-System interface Effective Exponential SINR mapping (EESIM) [15]DL Low traffic model Voice Over Internet Protocol (VoIP)
Silence/Talk Spurt length = exp. with 3 sec meanCall duration = 10 secInter-repetition time: uniform distributed ∈ [5, 10] sec
DL Peak traffic model Video conferencing applicationIn/Outgoing stream IAT = Const (0.01 sec)In/Outgoing stream frame size = Const (80 Bytes)Duration = Const (300 sec) (see Table 2)
Hypervisor resolution 1 sec (this is only for the dynamic case)Simulation run time 1000 sec
In order to show the effect and benefits of the wireless virtualization andthe air interface resource sharing, a peak traffic model is introduced to thesimulation scenarios. As it can be seen from the table above, the peak trafficmodel is used for only 300 seconds, this emulates a sudden peak in the load ofthe operator. The exact time where this peak traffic is used for each operatoris configured as follows:
Table 2 Peak traffic setup for each virtual operator
Cell1 Cell2 Cell3Virtual Operator 1 0 - 300 sec 600 - 900 sec 300 - 600 secVirtual Operator 2 300 - 600 sec 0 - 300 sec 600 - 900 secVirtual Operator 3 600 - 900 sec 300 - 600 sec 0 - 300 sec
In the static scenario three different operators are operating in the sameregion, each operator uses his own frequency band. These bands are being pre-allocated at the beginning of the simulations. Since we have 99 PRBs in total,
LTE Mobile Network Virtualization 9
each operator will get one third of the PRBs (that is 33 PRBs). The dynamicscenario can be looked at as the futuristic approach where the three operatorsare actually sharing the frequency band dynamically depending on their trafficload and requirements. At each second the hypervisor tries to calculate fromthe previous time instance what the traffic load of each operator is and howthe channel conditions experienced by the operators users look like, and thenassigns the PRBs among the different operators based on these calculations.
Fig. 3 Cell1 per virtual operator (VO) allocated bandwidth (MHz)
Figure 3 shows the bandwidth of each of the virtual operators that the hy-pervisor has allocated. It can be noticed that the allocated bandwidth changesover time depending on the traffic load and the users channel condition of eachoperator. In the first 300 seconds of the simulation run time it can be seenthat operator 1 has been allocated a much higher bandwidth compared to theother two operators. This is due to the scenario configuration (table 2), whereoperator 1 will have four additional users with video applications causing anincrease in their operator’s traffic load. Similarly there are additional users inthe second 300 seconds for virtual network 2 (VO2) and in the last 300 secondsfor virtual network 3 (VO3). The average per user air interface throughput foroperator 1 users in cell number 1 can be seen in figure 4. The upper part of thefigure shows the air interface throughput for the VoIP users whereas the lowerpart shows the video users throughputs. What can be noticed from these fig-ures is that the dynamic scenario achieves an expected better throughput thanthe static case. The reason for that is the additional resources that are used inthe dynamic case, where in the static case these resources were allocated forthe other operators but were not used.
10 Yasir Zaki et al.
10
15
20
25
Throughput (kbps)
Downlink VoIP per User Air Interface Throughput
Static
Dynamic
0
5
10
15
20
25
1 2 3 4 5 6 7 8 9 10
Throughput (kbps)
User number
Downlink VoIP per User Air Interface Throughput
Static
Dynamic
40
60
80
100
120
140
Throughput (kbps)
Downlink Video per User Air Interface Throughput
Static
Dynamic
0
20
40
60
80
100
120
140
1 2 3 4
Throughput (kbps)
User number
Downlink Video per User Air Interface Throughput
Static
Dynamic
Fig. 4 Cell1 virtual operator 1 per user air interface throughput (kbps)
The average user application end-to-end delays can be seen in figure 5. Itcan be noticed that the static scenario suffers from higher application delaysas compared to the dynamic case especially for the video users. The reason isthe limited air interface resources.
LTE Mobile Network Virtualization 11
40
60
80
100
120
140
160
Delay (ms)
Downlink VoIP per User End‐to‐End App. Delay
Static
Dynamic
0
20
40
60
80
100
120
140
160
1 2 3 4 5 6 7 8 9 10
Delay (ms)
User number
Downlink VoIP per User End‐to‐End App. Delay
Static
Dynamic
10000
15000
20000
25000
30000
Delay (ms)
Downlink Video per User End‐to‐End App. Delay
Static
Dynamic
0
5000
10000
15000
20000
25000
30000
1 2 3 4
Delay (ms)
User number
Downlink Video per User End‐to‐End App. Delay
Static
Dynamic
Fig. 5 Cell1 virtual operator 1 per user end-to-end application delay (ms)
In order to check how the VoIP application performs, the average delayvalues shown in the previous figures are not sufficient. So in addition figure6 shows the probability when the end-to-end delays of the VoIP packets weregreater than 300 ms (which is the QoS limit for the VoIP packets). It can beseen that the static case has a higher probability for not satisfying the QoSthreshold.
12 Yasir Zaki et al.
2.0%
3.0%
4.0%
5.0%
6.0%
7.0%
Probab
ility %
Probability that the VoIP End‐to‐End Delay > 300ms
Static
Dynamic
0.0%
1.0%
2.0%
3.0%
4.0%
5.0%
6.0%
7.0%
1 2 3 4 5 6 7 8 9 10
Probab
ility %
User number
Probability that the VoIP End‐to‐End Delay > 300ms
Static
Dynamic
Fig. 6 Cell1 virtual operator 1 end-to-end delay probability greater than 300 ms
As a summary, the simulation results show the advantage of using thedynamic spectrum sharing hypervisor. This can be seen in the better qualityof service of the user’s performance parameters, in particular for the videousers. In the dynamic case the video users do not suffer from high end-to-end delay as compared to the static scenario, although both scenarios use thesame total amount of air interface resources and the same configuration. Theonly difference is that in the dynamic case a better overall resource allocationhas been achieved and no resources have been wasted. Some of the results(especially the ones related to the VoIP users) showed that the static scenariohas a slightly better performance mainly when it comes to the end-to-enddelays. The reason for this is that in the static case there are not sufficientoverall resources to serve all of the video and VoIP users. Since the videotraffic was put in a lower MAC priority class, and due to not having enoughresources to fulfil all of the video user requirements in all of the TTIs, more freeresources were left available to serve the VoIP users which explain the betterperformance. Whereas in the dynamic case, there are sufficient resources toserve all of the video users, this means fewer resources for the VoIP users. Theresults of the other two virtual operators are not shown since they are similarto the virtual operator 1 results.
5.2 Multi-User Diversity Use Case
This use case investigates another gain aspect of the the wireless air inter-face virtualization process. Since LTE uses OFDMA, the MAC scheduler dis-tributes the air interface resources among the different users in both time andfrequency domain. The user channels are normally frequency selective, whichmeans each user will be experiencing different channel conditions on differentPRBs. From that the well known “Multi-user diversity” concept comes into
LTE Mobile Network Virtualization 13
the picture, where the channel-aware scheduler always tries to take advantageof the different channel conditions of the users to assign the PRBs to the usersexperiencing better channel conditions. Virtualization enables the virtual op-erators to share their individual spectrum, which means the scheduler has abigger set of PRBs to exploit the multi-user diversity. In summary, this usecase aims to show that even in the case where no multiplexing is allowed be-tween the different virtual operators and each operator still has a fixed amountof spectrum, a gain could be achieved if the spectrum is dynamically chosenbased on the user channel conditions. In this use case two different scenariosare compared against each other as follows:
– Static scenario: which is similar to today’s mobile network setup where eachoperator owns a fixed amount of spectrum that is pre-allocated throughoutthe whole simulation
– Virtual scenario: which aims to show how the multi-user diversity gaincould be achieved by sharing the spectrum among the different operators,keeping in mind that the total amount of spectrum each operator gets isfixed and not changing (no multiplexing)
Apart from the above mentioned difference and the configurations listed inTable 1, the rest of the simulation configurations are in Table 3. A full-bufferFTP traffic model and Max-C/I scheduler are selected for these simulationsin order to exploit the multi-user diversity gain.
Table 3 Simulation configurations
Parameter AssumptionNumber of virtual operators 3 virtual operators each with one eNBeNodeB coverage area Circular with radius = 375 meters with one cellTotal Number of PRBs - 75 PRBs or 15 MHz (5MHz for each operator)
- 150 PRBs or 30 MHz (10MHz for each operator)- 300 PRBs or 60 MHz (20MHz for each operator)
Number of active users VO1: 1, 2, 3, 4, 5, 10, 15 FTP usersVO2: 1, 2, 3, 4, 5, 10, 15 FTP usersVO3: 1, 2, 3, 4, 5, 10, 15 FTP users
Link-to-System interface Effective Exponential SINR mapping (EESIM) [15]DL traffic model FTP traffic with full buffer occupancyMAC layer scheduler Max-C/I
Figure 7 shows the average cell throughput achieved by virtual operator 1.The cell throughput is shown against various number of users (starting from1 user all the way up to 15 users) as well as different values for the systembandwidth (i.e. 5, 10 and 20 MHz). There are several aspects that can bepointed out from the results. First of all, the average cell throughput seen inthe figure is higher for the virtual scenario compared to the static case, inother words there is a gain in the cell throughput that can be achieved byapplying virtualization which comes solely from multi-user diversity. This isbecause the virtual operator can choose which PRBs to use to schedule hisusers from a larger spectrum selection (3 times larger in our investigation).
14 Yasir Zaki et al.
It could also be noticed that the gap between the virtual scenario and thestatic one starts to close when the overall number of users is increased. Thereason for that is the fact that when the number of users increase the multiuser diversity gain within the static scenario is nearly completely exploited,because the probability of finding a user with good channel condition on eachPRB increases.
20 0
30.0
40.0
50.0
60.0
70.0
80.0
ell throughput (M
bps)
VO1 Cell Throughput
static 5MHz
virtual 5MHz
static 10MHz
virtual 10MHz
t ti 20MH
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
1 2 3 4 5 10 15
Cell throughput (M
bps)
Number of Users
VO1 Cell Throughput
static 5MHz
virtual 5MHz
static 10MHz
virtual 10MHz
static 20MHz
virtual 20MHz
Fig. 7 Virtual Operator 1 cell throughput with and without virtualization
The absolute gain in cell throughput can be more clearly seen in figure 8.This is mainly the difference in cell throughput between the virtual and thestatic scenario seen in the previous figure. Again, it can be observed that thegain in cell throughput decrease when the number of users increase. Whatcan be additionally seen from the results that when the overall spectrum isincreased the gain in cell throughput increases as well, this is because theprobability of finding a PRB where the user is experiencing good channelconditions is much larger due to the larger spectrum selection. One thingshould be pointed out that for the results of the one user there is no multi-user diversity gain at all, and that is why the absolute gain increases fromthe points of 1 user to 2 users and then gradually goes down onwards byincreasing the number of users. The results of the other two virtual operatorsare not shown due to their similarity to operator 1.
LTE Mobile Network Virtualization 15
2.00
3.00
4.00
5.00
6.00
7.00
throughput Gain (Mbps)
VO1 Cell Throughput Gain due to Virtualization
5MHz
10MHz
20MHz
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
1 2 3 4 5 10 15
Cell throughput Gain (Mbps)
Number of Users
VO1 Cell Throughput Gain due to Virtualization
5MHz
10MHz
20MHz
Fig. 8 Virtual Operator 1 cell throughput gain due to virtualization
In summary, using virtualization enables the operators to share their spec-trum which leads to a number of advantages. In this use case, it was shownthat even though each operator still uses a fixed amount of spectrum (no mul-tiplexing) but being able to choose where the allocated spectrum lies can leadto having a better cell throughput. This gain comes from the multi user diver-sity effect, because the probability of finding good air interface resources foreach user increases compared to the static scenario.
6 Conclusion and Outlook
This paper shows the advantages introduced by the LTE virtualization basedon spectrum sharing between different virtual operators. Two different ad-vantages are analysed: multiplexing gain and multi-user diversity gain. Themultiplexing gain comes from the changing dynamics of the operator’s load.Since the traffic load of each virtual operator changes over time, the spectrumutilization will be different time to time. At some time instances one virtualoperator might have free resources which provides the opportunity to the othervirtual operators who need more resources to use this spare spectrum. In ad-dition, by sharing the frequency spectrum between the virtual operators, theprobability of finding nearly optimal resource allocation for each virtual oper-ator is higher in the frequency domain leading to the multi-user diversity gain.Such a gain is only obvious when the number of users in each virtual operatorsis low, and as soon as the number of users starts increasing, the effect of themulti-user diversity is decreased.
In addition to the above achievements, further advantages of LTE virtual-ization can also be seen: for example sharing the infrastructure of the LTE sys-tem is a considerably large gain because it leads to having less equipment andthus reducing the power consumption. It also reduces the capital investment
16 Yasir Zaki et al.
since the operator does not need to buy the equipment but only lease virtualresources. Another advantage of LTE virtualization is the flexibility the virtu-alization brings into the operator’s networks, the operator can shrink/expandand do changes to his virtual network on the fly very easily which is one ofthe major advantages of network virtualization.
Although many advantages of LTE virtualization can be observed throughthis paper, there are still many issues to be solved before having a commercialsolution of LTE virtualization. For example, designing a hypervisor is requiredto perform an efficient resource allocation in real time; the spectrum sharingshould be done in a cooperative way in a multi-cell scenario in which the in-terference management has to be considered. These issues will be investigatedin our future research work.
References
1. Niebert N, Baucke S, El-Khayat I, et al.: ”The way 4WARD to the creation of a FutureInternet”, in: ICT Mobile Summit, Stockholm, June 2008.
2. N. Feamster, L. Gao, J. Rexford; How to lease the Internet in your spare time; ACMSIGCOMM Computer Communications Review, Vol. 37, No.1, January 2007.
3. GENI Planning Group,”GENI: Conceptual Design, Project Execution Plan”, GENI De-sign Document 06-07, January 2006. http://www.geni.net/GDD/GDD-06-07.pdf
4. AKARI Architecture Conceptual Design for New Generation Network 1.1 http://akari-project.nict.go.jp/eng/concept-design/AKARI fulltext e translated version 1 1.pdf
5. Asia Future Internet (AsiaFI) (www.asiafi.net)6. J. Chapin; ”Overview of Vanu Software Radio”; from http://www.vanu.com, 20097. A. P. Subramanian, H. Gupta, S. R. Das, M. M. Buddhikot; ”Fast Spectrum Allocation in
Coordinated Dynamic Spectrum Access Based Cellular Networks”; DySPAN 2007, Dublin,Ireland, Apr. 2007.
8. J.A. Village, K.P. Worrall, D.I. Crawford; ”3G Shared Infrastructure”; 3G Mobile Com-munication Technologies, May, 2002
9. R. Agrawal, R. Berry, Huang Jianwei, V. Subramanian: Optimal Scheduling for OFDMASystems. Signals, Systems and Computers, 2006. ACSSC ’06. Fortieth Asilomar Confer-ence.
10. Rajiv Agarwal, Vinay Majjigi, Rath Vannithamby, John Cioffi, Efficient Scheduling forHeterogeneous Services in OFDMA Downlink, IEEE Globecom 2007, Washington D.C
11. M. Einhaus, O.Klein, Performance Evaluation of a Basic OFDMA Scheduling Algorithmfor Packet Data Transmissions, ISCC 2006, Cagliary, Italy
12. Michael Einhaus, Ole Klein and Bernhard Walke, ”Comparison of OFDMA ResourceScheduling Strategies with Fair Allocation of Capacity”, 2008 5th IEEE Consumer Com-munications & Networking Conference (CCNC 2008), Las Vegas, NV, January 2008
13. OPNET website: http://www.opnet.com14. M. Anas, F.D Calabrese, P.E Mogensen, C. Rosa, K.I Pedersen: Performance Evaluation
of Received Signal Strength Based Hard Handover for UTRAN LTE. Vehicular TechnologyConference, 2007. VTC2007-Spring IEEE 65th.
15. Erik Westman, Calibration and Evaluation of the Exponential Effective SINR Mapping(EESM) in 802.16, Master Thesis, Stockholm, Sweden, 2006.
