The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks

Bring Your Own Network – Design and Implementation of a Virtualized WiFi Network – ∗1

Kiyohide NAKAUCHI∗1 Yozo SHOJI∗1 ∗1 ZHONG Lei Yoshinori KITATSUJI∗2

Manabu ITO∗1 Hidetoshi YOKOTA∗2

National Institute of Information and Communications Technology, Tokyo, Japan {nakauchi, shoji, mn-ito, zhong}@nict.go.jp ∗2 KDDI R&D Laboratories, Inc., Saitama, Japan {kitaji, yokota}@kddilabs.jp

Abstract— This paper proposes virtualized WiFi network that can dynamically create a virtual Base Station (vBS) around the target mobile devices that offers dedicated base station resources for satisfying service-specific QoS and works as a gateway for a corresponding virtual network or slice. Specifically, the paper proposes (1) a technique to dynamically configure a vBS on top of multiple physical WiFi base stations by exploiting the features of OpenFlow, and (2) a technique of network-driven seamless handover between vBSs by forced association and authentication in advance at a target vBS. The paper also describes a detailed design and implementation of a physical WiFi base station which can organize a vBS, named virtualization capable WiFi Base Station (vcBS). As a prototype, two vcBSs and virtualization capable Base Stations Switch (vcBS-SW) to accommodate and centrally control those vcBSs are newly developed. The paper demonstrates a vBS can be dynamically configured on top of two vcBSs and the base station resource can be dynamically allocated to the vBS by assigning additional WiFi interfaces or frequency channels based on a resource allocation policy. The paper also demonstrates the proportion of SIP calls whose setup time exceed the threshold of 600ms can be reduced from 19.7% to 4.6%, when the SIP signaling traffic is served by a SIP-specific vBS. Finally the paper demonstrates that the prototype system can make seamless handover for a target device from common vBS to service-specific vBS in less than 65 ms without any packet drop. Keywords— Network Virtualization, Virtual Base Station, Mobile Network, Wireless LAN

I. I NTRODUCTION Considering recent advances of wireless technologies, there is a strong demand to cope with the increasingly diversifying requirements of emerging mobile services provided over mobile networks as well as short-term dynamism of requirements caused by user mobility and service user distribution accordingly. However, the introduction of new functionality to meet those requirements into wireless access networks is prevented by economic considerations. Apart from the costs implied by the development process, the infrastructure costs of a wireless access network with base station sites, antennas, radio control nodes and backhaul links is one of the largest cost components within a cellular network [8]. Any build-out of the radio access network with new functionality requires long-term planning and large investment in general. Network virtualization is a promising way to lower such economic barrier for the deployment of new technologies and service-specific customization to flexibly accommodate the diversity and the dynamism on a shared infrastructure [2], [4], 978-1-4799-2355-7/14/$31.00 ©2014IEEE

[9], where independent virtual networks (slices) with servicespecific configurations coexist. Recent technical advance of Software-Defined Network (SDN) is expected to make network virtualization a reality. At the same time, those concepts are extended to mobile networks, and architectures of mobile network virtualization or software-defined mobile network are discussed as a key solution for mobile operators to handle short-term and long-term dynamism of resource demands in mobile networks. The difficulties of creating such end-to-end slice over wireless and wired networks are essentially to handle not only the dynamic characteristics of radio and interferences but also the dynamic characteristics of user distribution due to mobility.There are several work on virtualization of wireless base stations [5], [7], [11]. Designs of control plane based on the principle of data and control plane separation for operating those virtualized base stations are also discussed. For example, OpenRadio [3] is a novel design for a PHY/MAC-level programmable wireless dataplane based on the unique combination of the programmability-performance-price metrics. CellSDN [6] is a software-defined cellular network architecture that support dynamic traffic control policy, flexible slicing of network resources based on the attributes of subscribers, and flexible slicing of base stations and wireless resources. However, it is not well considered how to isolate traffic of a specific service such as delay-sensitive VoIP service even in the condition of congestions for mobile users where a huge number of terminals coexist and a huge number of base stations are densely arranged. This paper proposes virtualized WiFi network that can dynamically create a virtual Base Station (vBS) around the target mobile devices that offers dedicated base station resources for satisfying service-specific QoS and works as a gateway for a corresponding virtual network or slice. Specifically, the paper describes a detailed design of vBS, which can configure ”virtual coverage” for a specific service, and shows a design of a seamless handover between vBSs. A WiFi-based prototype system that consists of a set of virtualization capable Base Stations (vcBSs) hosting vBSs and virtualization capable Base Stations Switch (vcBS-SW) accommodating those vcBSs and centrally controlling those vcBSs is also shown. The organization of this paper is as follows. In Section II, the basic concept of virtualized WiFi networks is proposed.In Section III, a technique to dynamically configure vBS in a logical integration manner and a seamless inter-vBS (interslice) handover technique are proposed. Section IV describes a WiFi-based prototype system which consists of two vcBSs and vcBS-SW. Section V shows an experimental system

483

The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks

and preliminary experimental results. Finally, we conclude in Section VI. II. V IRTUALIZED W I F I N ETWORKS A. Bing Your Own Network (BYON) and Its Requirements for Wireless Access Networks Bring Your Own Network (BYON) is our research conceptual goal, which enables dynamic configuration of a servicespecific virtual network (slice) around service devices by dynamically migrates and localizes related service resources as well as configuring a service-specific wired and wireless network in a coordinated manner considering the distribution shape of the devices using the specific-service as well as their mobility tendency [10]. Quick response in application, higher efficiency in resource utilization, and reduction of signaling messages are consequently expected by dynamically configuring a service-specific mobile network that satisfies required QoS even in the condition of congestions for mobile users, keeps continuous service accessibility with the same device configurations, and also securely manages association and authentication among base stations. In BYON, two types of slices are defined, and each mobile device is connected to one of them based on the service the device joins. ”Common slice” is an open slice for every service devices, and accepts accesses by any type of devices to associate with it. A common slice deals with control packets mainly but also data packets and offers only best-effort quality of service. The common slice is used for service initiation for every device. The other is ”service-specific slice”, which is invisible and inaccessible except for the eligible service devices. In the service-specific slice, QoS, efficiency in resource utilization, security, and usability are optimized for a specific service. The service-specific slice is dynamically and strategically configured only around the locations where service demands exist. The service-specific slice is moved or expanded, corresponding to the changes of the distribution shape of the service users as well as their mobility tendency. For the slice operation in BYON described above, the following requirements should be satisfied in wireless access networks. • Dynamic slice configurations: To dynamically configure service-specific slices around mobile users with required QoS must be supported. Per-flow control is also required to support flow-level mapping to slices. Dynamic resource assignment to the service-specific vBSs should be also supported. • Smooth transition between slices: Inter-slice seamless handover from a common slice to a service-specific slice should be supported to prevent disruption of the target services. As a service can be defined as a set of flows in a flow level, inter-slice handover should be managed on a per-flow basis. Here we identify a flow by a subset of source and destination fields in MAC and IP header, and port number field in transport header. • Dynamic slice expansion: To dynamically and continuously move or expand service-specific slices around the users should be supported when the locations of users or service access frequency change. From a user perspective, mobility should be hidden. That means wireless access networks should enable users to be always connected to the same base station logically even when those users are physically connected to different base stations in the same administrative domain or across administrative domains.

 



 





   
 




   





 



 


 
 

 
 


 !   

 
 
  
 !   

 
 



Fig. 1.

Network Model of Virtualized WiFi Network

B. Network Model of Virtualized WiFi Network Figure 1 shows a basic network model of virtualized WiFi network. Virtualized WiFi network is a logical WiFi network that is built on top of physical WiFi networks regardless of physical structure of them. Virtualized WiFi network hosts its own reconfigurable software virtual WiFi BS (vBS). Virtualized WiFi network statically provides a common slice and dynamically host service-specific slices on an on-demand basis. Common vBS and service-specific vBS can be considered as an entrance point to a common slice and service-specific slide for the service users, respectively. The important design goal of the virtualized WiFi network is mobility support. We consider two aspects of mobility support in virtualized WiFi network. The first aspect is handover between vBSs or slices. Mobile users of a specific service that requires a certain wireless access resource make seamless handover from a common vBS to a service-specific vBS. The second aspect is mobility of vBS itself. When the users physically or logically move to another BS’s coverage, the service-specific slice is expanded or moved to accommodate the BS. At that time, terminal state information of association and authentication for the target terminals are shared among BSs belonging to the same vBS. Consequently the users can always be connected to the same vBS then the uses are in the corresponding BS’s coverage. That’s the reason why those users recognize that the service-specific vBS follow the users when they move. In summary, the essential features of virtualized WiFi network (the ways how to meet the requirements of BYON described in Sec. II-A) are as follows. • Dynamic slice configurations is satisfied by configuring logical integration based vBS in a software on top of vcBSs with dedicated wireless access resource. • Smooth transition between slices is satisfied by seamless inter-vBS (inter-slice) handover. • Dynamic slice expansion is satisfied by vBS mobility while keeping terminal state information of association and authentication. It is our future work to address interdomain vBS mobility. As shown in Figure 1, virtualized WiFi network is composed of a set of virtualization capable WiFi Base Stations (vcBSs) and virtualization capable Base Stations Switch (vcBS-SW). vcBS is a physical BS that hosts common and servicespecific vBSs. We assume multiple vcBS are deployed in the

484



The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks

 "#'
 $
!# 


''

+*$'%1.0'+*/


 "#'
 
"% 


4
4
4


)#
 
"".#//
)#

)#



 "#

$


&!

 
"% !


Fig. 2.


 !# ('

 "#
 
$  


 "#'
 
"% 


 "#
$ 


$ 
!


2

 "#'"


$


&!
 
 "




3(*
&




 

.+ #

#-1#/0


Logical Model of Virtualized WiFi Network





 !# ('




2

3(*
& 




.+ #

#-1#/0










3(*
&






.+ #

#-1#/0












3(*
&

.+ #

#-1#/0







.+ #

#/,+*/#




 
('!#


same location or in distinct locations. Typical examples are Enterprise WiFi and WiFi hotspot, respectively. The principle of network model of virtualized WiFi network is centralized control of base station resource management, association, authentication, and handover. This principle is needed to configure a well-managed vBS among multiple physical vcBSs. vcBS-SW is located on the data path and accommodates vcBSs.Note that this paper focuses on a case of WiFi networks, though conceptually BYON does not depend on a specific wireless access technology.

Fig. 3. Configurations of Wireless Interfaces and Coordinated Handling for Probe Request Frames

III. D ESIGN D ETAILS OF V IRTUALIZED W I F I N ETWORK

C. Logical Model of Virtualized WiFi Network

In this paper, we propose a technique to dynamically configure vBS in a logical integration manner and a seamless intervBS (inter-slice) handover technique. Though a vBS mobility mechanism for dynamic slice expansion is also needed for BYON, which is our current work and is not mentioned in this paper. However, we believe the essential part is shared with the inter-vBS (inter-slice) handover technique.

In general, there are two approaches to configure vBS. The first is a logical separation approach, where single WI/F is logically separated into multiple vW-I/Fs. However, configuring logical separation based vBS aims at creating a slice with dedicated wireless resource and privacy on the physical BS, and does not consider mobility support and smooth transition between slices, which are the requirements of BYON. In this paper, we focus and take a logical integration approach, because a slice that covers multiple physical BSs needs to be considered in BYON. Here, logical integration means that multiple physical BSs are logically integrated into a single vBS and multiple BSs cooperatively behaves as a single logical BS. Figure 2 shows a logical model of virtualized WiFi network based on the logical integration approach. More importantly, in each slice, one or multiple vBSs are configured. Each vBS has its own vW-I/F, and this means vBS can be configured independent of physical structure of a WiFi network. From a user’s perspective, a slice is visible as vBS or vW-I/F. Thus, in the virtualized WiFi network, association with vBS means that a device is associated with one of the physical BSs that composes the vBS. Figure 2 also shows a relationship between vBS, virtual wireless Interface (vW-I/F), virtual coverage, and a slice. In this model, we comparably use the words ”vBS” and ”vW-I/F”, and ”BS” and ”W-I/F”, respectively. Each vBS has its own virtual coverage, and single or multiple cells of corresponding physical vcBSs organize the virtual coverage. Note that a design goal of the virtualized WiFi network is to configure vBS on top of vcBSs whose cells are completely separated. If those vcBSs are connected to different vcBS-SW, it is more challenging because some coordination between vcBS-SWs are needed. In this paper, we discuss on the configuration of vBS on top of vcBSs whose cells are overlapped, because we should discuss how to realize inter-vBS (inter-slice) seamless handover for BYON.

A. Configuring Virtual Base Station (vBS) This section proposes a technique to dynamically configure vBS on top of multiple vcBSs in a WiFi network. Figure 3 shows how to configure vcBSs to host vBSs. In this example configuration, vBS#1 for a common slice is created by two W-I/Fs (wlan0, wlan1), and vBS#2 for a servicespecific slice is also created by two W-I/Fs (wlan2, wlan3). In order to make two W-I/Fs behave as a single W-I/F in an coordinated manner and to realize smooth transition between vBS#1 and vBS#2, all the W-I/Fs (wlan0-3) are configured with the same MAC address. In addition, same ESSID and same BSSID are configured in WiFi AP software. Note that it does not matter whether all the W-I/Fs (wlan0-3) are equipped in a vcBS (multi-interface BS) or each W-I/F is equipped in a different vcBS. As a result of the above W-I/F configurations, all the beacon frames transmitted from those W-I/Fs have the same source MAC address and BSSID in the frame. Only the channel information is different. In order to avoid terminals to remember the channel for corresponding ESSID, vcBSs are operated by stealth mode (the ESSID field in a beacon is empty). Regardless of slices, all the terminals recognize only one BS with dynamic channel selection. Then another mechanism is needed to determine which terminal is associated with which W-I/F. Figure 3 also shows how vcBSs behave against probe request frames. The mechanism of finding available BSs using probe request and probe response is a basic function in IEEE 802.11. The above vcBS configurations cause terminals transmit prove request frames for target ESSID using different channels one by one. If the terminal (STA#2) is belonging to Slice#2, vBS#2 responds to the request. As vBS#2 has two available W-I/Fs (wlan2, wlan3), the W-I/F with less number of associations (wlan3) is chosen. Then vcBS that equipped wlan3 responds to STA#2’s

485

The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks


(!1)-'
9 
 :
!,.-'
5 2
)-
!$5!-#%



 

 

 
.1#%$
$% !43(%-3)#!3).-
&.1
3!1'%3
3%1,)-!+2




  





 
!23
1%1.43%
42)-'
/%-+.6



5#



%6
#(!--%+
)2
-.3)&)%$
3.
%!#(
42%1
42)-'

(!--%+
6)3#(
--.4-#%,%-3
)-
!
1."%
%2/.-2%
&1!,%


 





 
%,/.1!+
"4&&%1)-'
&.1
3!1'%3
&+.62
)-




.1#%$
/1% !43(%-3)#!3).-!22.#)!3).-
6.
!-7
#.,,4-)#!3).-
6
2


/%-+.6

5#


5#
/%-+.6

/%-+.6

%1,)-!+
23!3%


Fig. 4.

1."%
%2/.-2%
6





vcBS Basic Design with Multiple vBSs

5 


5 


))


))


B. Seamless Inter-vBS Handover This section proposes a technique to make seamless handover from a common vBS to a service-specific vBS for target service flows. For simplicity, we assume each terminal has a primary flow. When vcBS chooses the primary flow as a handover target, the terminal makes handover from a common vBS to a service-specific one. Figure 5 shows a seamless inter-vBS handover mechanism. vBS#1 is hosted only by vcBS#1, and vBS#2 is hosted by both vcBS#1 and vcBS#2. Specifically, vBS#2 is created on top of WiFi Interface #4 (WiFi#4) in vcBS#1 and WiFi#1 and WiFi#2 in vcBS#2. All the active WiFi Interfaces are operated with

Fig. 5. One

))


))










5 
 ))


))


))


))






1."%
%2/.-2%
6













probe request. In our model, vcBS-SW is responsible for managing and determining these associations. vcBS-SW is the key element from the perspective of control plane of virtualized WiFi network. The main roles of vcBSSW are as follows. • To obtain service descriptions (source/destination IP addresses, port numbers, service users’ IP addresses, etc.) • To manage the mapping among a service, a flow, and MAC address of a terminal corresponding to the flow • To identify flows based on source and destination IP address fields • To determine creating or shutting down of vBSs • To control wireless access resources allocated to vBSs • To manage and determine associations between terminals and W-I/Fs • To determine the timing and target flows for inter-vBS handover. Figure 4 shows a basic functional design of vcBS. vcBS consists of flow control part, BS resource abstraction layer, and BS resource stack. First, flow control part is responsible for switching downstream flows among vBSs and switching upstream flows among virtual wired networks. Because both dynamic slice configuration and per-flow control are required as described in Sec. sec:vWiFi-Req, we adopt OpenFlow due to its high flexibility. Second, BS resource abstraction layer is responsible for logical integration of W-I/Fs and switching among W-I/Fs for downstream flows. BS resource abstraction layer is implemented by software. The resource boundary that separates vBSs is dynamically adjusted by increasing or decreasing the number of W-I/Fs assigned to those vBSs. Finally, BS resource stack is a pool of physical wireless resources. In this figure, vcBS is equipped with four W-I/Fs and BS resource stack consists of at most four independent frequency channels. This means vcBS can host at most four vBSs. The detailed design of vcBS is shown later in Sec. IV.

5


5


8

!)1
1!-2)%-3
%7

8

!#*%3
4,"%1
8

%04%-#%
.4-3%1
8
!/!")+)3)%2



(72)#!+
)1%+%22

5 
)134!+
)1%+%22














%1,)-!+
3!3%


Procedure of Handover from a Common Slice to a Service-specific

different channel. In this figure, both STA#2 and STA#3 make inter-vBS handover from vBS#1 to vBS#2, while STA#1 stays at vBS#1. Specifically, STA#3 makes handover from WiFi#1 in vBS#1 to WiFi#1 in vBS#2. The proposed seamless inter-vBS handover technique is original from the following perspectives. • Sharing ”terminal state” among vBSs in advance, and forced pre-authentication and pre-authentication based on the share terminal state without any communication with target STAs • Forced transmission of probe response frame by unicast only to the target terminals, what notifies the new channel • Temporal buffering of target downstream flows using OpenFlow at vcBS-SW to prevent frame drops during handover • Forced rerouting of target downstream flows using OpenFlow both at vcBS-SW and vBS to achieve fast handover In the case of WPA2 CCMP, terminal state is defined as a set of CCMP Pair Transient Key (PTK), CCMP Packet Number, WPA Sequence Counter, and Capabilities. CCMP PTK is needed to skip 4-way handshake procedure to create the key. CCMP Packet Number and WPA Sequence Counter are needed for synchronizing sequence number of transmitted data frames among W-I/Fs during handover. Capabilities information is also needed to skip association procedures. After completing forced pre-authentication and preauthentication based on the shared terminal states of STA#2 and STA#3 in Figure 5, new channel 40 and 44 are notified to STA#2 and STA#3, respectively. If a terminal need not change frequency channel for handover, this step can be skipped. The CSA (Channel Switch Announcement) field in a probe response frame is used for the channel notification. Probe response frames are suitable for the purpose because vcBS can send these probe response frames only to the target terminals by unicast. At that time, vcBS-SW temporally buffers target flows during handover. OpenFlow’s packet-in mechanism is used for this purpose. Specifically, the corresponding flow entries are immediately removed when registered. In addition, OpenFlow switch is forced not to response to packet-in in order to buffer every packet of target flows. It should be emphasized that the proposed inter-vBS handover technique does not require any software modification of WiFi terminals. Only the IEEE 802.11’s standard frames and

486

The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks

#'-0


#'-0


) '
./#," ,"




* ,

 .# +





,"

) '%(/


* ,

 .# +





,"





*














,"
 */

* ,
(.+
*#! +


.


.*





*

.*


*





.


.*


*

,"
 */











.


,"








) '
./#,"

* )*( ++#'!

%(/
(',*(%


Fig. 6.

(+,)


) '%(/


) '
./#,"





,"
 */




* &

(',(%% *


.


,"





) '
./#,"





," */

* ,
.+


'.01 ('5$#

,$*(+3
3'0"&

(',*(%
-," ++(


.+(',*(%% *


.++/#,"


2" /%'




/%'




/%'




/%'





$.2$./
+))+*$.2'"$ /,$"'%'" 2"
,$"'%'" 0'+*

2"




#'

. 
'

 +(-*
,$




 

*


.$-
! *#

5



+%

)+#1($


1(0',($4$.

Software Architecture in vcBS

1(0',($4$.

protocols are used. All the intelligence are put into vcBS-SW, and no intelligence in terminals is required. IV. P ROTOTYPING We developed a prototype of virtualized WiFi network, which consists of one vcBS-SW and two vcBS to demonstrate its positive effect. This section shows the details of vcBS-SW and vcBS prototype, including software architecture of vcBS. The experimental system that consists of this prototype, WiFi terminal emulators, and SIP servers is described in Sec. V. A. Prototype System The vcBS-SW prototype is organized by production hardware OpenFlow-capable Layer 2 switch and x86 server for OpenFlow controller directly connected to the OpenFlow switch by a 10 GbE link. The OpenFlwo switch has 48x GbE and 4x 10GbE data ports, and can hold 1,000 flow entries as a specification. The OpenFlwo switch supports OpenFlow 1.0. We developed OpenFlow controller logic on Trema [1], an open source OpenFlow controller platform. All the intelligent mechanisms are also developed on the x86 server. On the other hand, vcBS prototype is composed of a preprocessing PC and a main PC, connected by a GbE link. The pre-processing PC hosts the OpenFlow part described in Figure 4. This embedded PC with an ARM processor equipped 4x GbE ports and software OpenFlow switch named Open vSwitch 1.7.1 (OpenFlow 1.2 is supported) is installed. On the other hand, the main PC provides BS resource abstraction layer and BS resource stack. Each vcBS equipped four IEEE 802.11 a/b/g/n 2.4GHz/5GHz dual-band modules (2x2 MIMO is supported) as the BS resource stack. The offthe-shell WiFi module equipped Atheros AR9280 chipset. We can operate those vcBSs with 2.4GHz and 5GHs (only W52; Channel 36,40,44,48) frequency band. vcBSs can also work as a normal multi-interface WiFi BS. Open vSwitch 1.7.1 is also installed in the main PC for flow switching among those W-I/Fs. The details are described in the next section.

''
. %%'"
$*$. 0+.


$.)'* (
)1( 0+.

Fig. 7.

'.$($//

$.)'* (
)1( 0+.

Experimental Setup to evaluate BYON

main PC. This OVS is responsible for switching flows among vW-I/Fs based on flow-slice mapping information. We also developed OpenFlow controller logic for those two OVS on Trema platform in the main PC. The OpenFlow controller is shared by OVS in the pre-processing PC. Flow entries for those OpenFlow switches and flow-slice mapping tables are registered by vcBS via control channel between vcBS and vcBS-SW, based on the flow identification mechanism in vcBS-SW. In order to maintain the slice logic defined by OVS in the pre-processing PC, VLANs are setup for each slice between the pre-processing PC and the main PC. Those VLANs are directly connected to corresponding vW-I/F in the main PC. By porting the intelligent mechanisms originally developed for vcBS-SW to this vcBS (main PC), we can operate the vcBS in a stand-alone mode without vcBS-SW.

V. P RELIMINARY E XPERIMENTS AND E VALUATIONS A. Experimental Setup Figure 7 shows an experimental system that we bulit to demonstrate and evaluate the effects of virtualized WiFi network based on the prototype system. In particular, we evaluated the effects in a typical VoIP over WiFi scenario. In the experimental system, a wireless SIP terminal emulator and SIP servers run SIP signaling through vcBS and vcBS-SW. A WiFi terminal emulator that can emulate a large number of WiFi terminals with indepedent MAC states is also introduced to generate cross traffic including Video, Audio, FTP, and HTTP services. Those emulators are connected to two vcBSs using RF cables, power dividers, couplers, and attenuators. We emulated virtualized wired networks using another production OpenFlow switch. B. Software Implementation in vcBS Those SIP servers have the capability of SIP session state Fig.6 shows a software architecture of vcBS. In the main migration from a common SIP server to a service-specific SIP PC, we develop a vW-I/F software module. The vW-I/F server. This experimental system can work for evaluating the software module is organize by TUN/TAP，vW-I/F process， performance of BYON by synchronizing inter-vBS handover and Open vSwitch (OVS) bridge. OVS here is responsible for and SIP session state migration for the target SIP signaling switching flows among W-I/Fs based on an association policy. flows. We modified hostapd 1.0 (open source WiFI BS software) for B. Reduction of SIP Call Setup Time cooperative association and authentication with vcBS-SW. On the other hand, in the pre-processing PC, another OVS In this experiments, SIP call setup time is measured. SIP bridge module is installed to offload flow processing from the calls are generated at the rate of 25 calls per second. SIP

487

The 11th Annual IEEE CCNC - Smart Spaces and Wireless Networks



ten times. As a preliminary result, handover successfully completed in less than 65 ms. Note that no packet drop is detected during handover due to temporal buffering of target downstream flows using OpenFlow at vcBS-SW.

 "



! #

 

 "



! #

 


 



Fig. 8.

Histogram of SIP Call Setup Time

call setup time for each SIP call is measured at vcBS-SW by detecting SIP INVITE message and 200 OK messages for each SIP session. In the experiment, the mixed traffic is generated at the rate of 21.16Mbps in total by fifty emulated terminals and three real WiFi devices. Three WiFi interfaces at a vcBS are activated with IEEE 802.11n 1x1 MIMO mode in the 5 GHz band. When vBS function is disabled and those three WiFi interfaces act as three independent WiFi base stations, cross traffic and SIP signaling traffic are accomodate by those WiFi interfaces randomly. When the vBS function is enabled, common vBS, Video-specific vBS, and SIP-specific vBS are created. Each vBS is assigned one physical WiFi interface in our configuration. This means dedicated base station resource, i.e. frequency channel, is assigned to the video service and the SIP service, respectively. Fig.8 shows the histogram of SIP call setup time for both cased. This measurement result demonstrates that by enabling vBS function at vcBS and vcBS-SW, the proportion of SIP calls whose setup time exceed the threshold of 600ms can be reduced from 19.7% to 4.6%, when the SIP signaling traffic is served by a SIP-specific vBS. C. Delay Performance of Seamless Inter-vBS Handover Finally, we validated the prototype system can achieve seamless inter-vBS handover. We configured two vBSs for this experiment. One is for common slice and wlan0 of vcBS#1 with channel 36 is assigned, and the other is for a service-specific slice and wlan0 of vcBS#2 with channel 44 is assigned. We used command line interface of vcBS-SW to manually initiate handover from common vBS to servicespecific vBS. In this experiment, a server connected to the production OpenFlow switch that emulates virtualized wired networks periodically generated packets with 60Byte payload at every 3 ms and sends them to a WiFi terminal (laptop PC equipped with the same WiFi module as vcBS). We measured the handover delay by identify the last frame from vcBS#1 and the first frame from vcBS#2. We conducted inter-vBS handover

VI. C ONCLUSIONS This paper proposed virtualized WiFi network that can dynamically create a virtual Base Station (vBS) around the target mobile devices that offers dedicated base station resources for satisfying service-specific QoS and works as a gateway for a corresponding virtual network or slice. Specifically, the paper proposed (1) a technique to dynamically configure a vBS on top of multiple physical WiFi base stations by exploiting the features of OpenFlow, and (2) a technique of networkdriven seamless handover between vBSs by forced association and authentication in advance at a target vBS. The paper also described a detailed design and implementation of a physical WiFi base station which can organize a vBS, named virtualization capable WiFi Base Station (vcBS). The paper demonstrated a vBS can be dynamically configured on top of two vcBSs and the base station resource can be dynamically allocated to the vBS by assigning additional WiFi interfaces or frequency channels based on a resource allocation policy. The paper also demonstrated the proportion of SIP calls whose setup time exceed the threshold of 600ms can be reduced from 19.7% to 4.6%, when the SIP signaling traffic is served by a SIP-specific vBS. Finally the paper demonstrated that the prototype system can make seamless handover for a target device from common vBS to service-specific vBS in less than 65 ms without any packet drop. R EFERENCES [1] Trema. http://trema.github.io/trema/. [2] T. Anderson, L. Peterson, S. Shenker, and J. Turner. Overcoming the Internet Impasse through Virtualization. IEEE Computer, 38(4), April 2005. [3] M. Bansal, J. Mehlman, S. Katti, and P. Levis. OpenRadio: a Programmable Wireless Dataplane. Proc. ACM HotSDN’12, August 2012. [4] A. Bavier, N. Feamster, M. Huang, L. Peterson, and J. Rexford. In Vini Veritas: Realistic and Controlled Network Experimentation. Proc. ACM SIGCOMM ’06, September 2006. [5] G. Bhanage, D. Vete, I. Seskar, and D. Raychaudhuri. SplitAP: Leveraging Wireless Network Virtualization For Flexible Sharing Of WLANs. Proc. IEEE GLOBECOM ’10, December 2010. [6] X. Jin, L. E. Li, L. Vanbever, and J. Rexford. SoftCell: Scalable and Flexible Cellular Core Network Architecture. Proc. ACM CoNEXT’13, December 2013. [7] S. Paul, S. Ganu, P. Kamat, and E. B. Royer. Requirements for Wireless GENI Experiment Control and Management. GENI Design Document 07-43, GENI Wireless Working Group, Februrary 2007. [8] J. Sachs and S. Baucke. Virtual Radio – A Framework for Configurable Radio Networks. Proc. WICON ’08, November 2008. [9] G. Schaffrath, C. Werle, P. Papadimitriou, A. Feldmann, R. Bless, A. Greenhalgh, A. Wundsam, M. Kind, O. Maennel, and L. Mathy. Network Virtualization Architecture: Proposal and Initial Prototype. Proc. ACM VISA ’09, August 2009. [10] Y. Shoji, M. Ito, K. Nakauchi, Z. Lei, Y. Kitatsuji, and H. Yokota. Bring Your Own Network – A Network Management Technique to Mitigate the Impact of Signaling Traffic on Network Resource Utilization –. Proc. MobiWorld 2014, January 2014. [11] G. Smith, A. Chaturvedi, A. Mishra, and S. Banerjee. Wireless Virtualization on Commodity 802.11 Hardware. Proc. WiNTECH ’07, September 2007.

488