software deﬁned networking in wireless mobile...

Software Defined Networking in Wireless MobileNetworks

Martin Nagy∗

Institute of Computer Engineering and Applied InformaticsFaculty of Informatics and Information Technologies

Slovak University of Technology in BratislavaIlkovicova 2, 842 16 Bratislava, Slovakia

[email protected]

AbstractThesis focuses on the topic of Software Defined Network-ing (SDN) in context of 3GPP (Third Generation Part-nership Project) track mobile networks. Software definednetworking is a trend that is slowly making its way tomost areas of computer networking. Nowadays it is visi-ble mainly in datacenter networking, private clouds and5G. However other technologies, such as 2G, 3G, LTE orWi-Fi can benefit from it as well. We proposed a new ar-chitecture for GPRS (General Packet Radio Service) de-livery (packet based 2G data). Architecture is designed tobe backwards compatible with the radio access networkwhich shall simplify the deployment. 2G may seem as anobsolete technology, however it is still widely used world-wide, thanks to its maturity and low production costs.Moreover it is a great fit for some of today’s IoT (Internetof Things) use cases. Integral part of the new architec-ture is a new tunneling approach called MAC tunneling,which replaces GTP (GPRS Tunneling Protocol) tunne-ling. The architecture is deployable not only with GPRS,but is generalized to be used with any other access tech-nology. Solution proof of concept is built and tested withstandard, not-modified 2G base station and terminal inorder to practically evaluate the architecture.

Categories and Subject DescriptorsC.2.1 [Network Architecture and Design]: Wirelesscommunication; C.2.3 [Network Operations]: Networkmanagement

KeywordsGPRS, SDN, NFV, Network Functions Virtualization, Soft-ware Defined Networking, UnifyCore, OpenFlow, mobilenetworks, 3GPP networks, wireless networks

∗Recommended by thesis supervisor: Prof. Ivan KotuliakTo be defended at Faculty of Informatics and Informa-tion Technologies, Slovak University of Technology inBratislava on April 29, 2019.

c© Copyright 2019. All rights reserved. Permission to make digitalor hard copies of part or all of this work for personal or classroom useis granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies show this notice onthe first page or initial screen of a display along with the full citation.Copyrights for components of this work owned by others than ACMmust be honored. Abstracting with credit is permitted. To copy other-wise, to republish, to post on servers, to redistribute to lists, or to useany component of this work in other works requires prior specific per-mission and/or a fee. Permissions may be requested from STU Press,Vazovova 5, 811 07 Bratislava, Slovakia.

1. IntroductionSDN (Software Defined Networking) is mostly visible inareas of fixed networks, mainly core networks and data-center networking to name a few. However, in the areaof cellular mobile networks, the SDN is just getting trac-tion, mostly thanks to the advent of 5G networks, whichpromise high utilization of SDN related technologies, forexample to provide network slicing [8]. While first stan-dard 5G networks are nowadays being deployed, our thesisfocuses on legacy cellular networking - networks which arealready widely deployed and used by the majority of thepopulation.

Our SDN approach was developed on the basis of 2G datanetworks (GPRS aAS General Packet Radio Service) andlater extended/generalized to other, even non-cellular ac-cess technologies.

OpenFlow protocol was used to bring SDN capabilitiesto our concept. We chose this protocol since it is open,well standardized and moreover, plethora of open-sourceOpenFlow implementations exist. This enabled practicalexperiments and evaluation of the whole concept.

2. Mobile NetworksGPRS network is an add-on to the ubiquitous 2G voicenetwork (GSM aAS Global System for Mobile Communi-cations). Voice services were not the focus are of our the-sis, therefore we will omit them going forward with thisdo-cument. For further information on the GSM archi-tecture and technologies, we encourage reader to studypublicly available 3GPP (Third Generation PartnershipProject) standards [5].

The 2G packet network added two new elements to thecore network of already existing GSM network and re-quired minor, mostly software changes in the radio accessnetwork [3, 2]. These nodes are SGSN (Serving GPRSSupport Node) and GGSN (Gateway GPRS Support No-de). SGSN is responsible for mobility and session ma-nagement of the connected mobile stations / devices (MS).These include procedures such as authentication, locationtracking, even traffic encryption and decryption. SGSNdirectly interfaces with the radio access network (RAN)by Gb interface, which connects SGSN and BSC (BaseStation Controller), the brain of the 2G access network.

GGSN connects to the SGSN on one side and to the PDNs(Packet Data Networks) on the other side. These can befor example various corporate intranets, public Internet

2 Nagy, M.: Software Defined Networking in Wireless Mobile Networks

Figure 1: Standard 3GPP GSM/GPRS architec-ture [2].

or even various dedicated application servers (e.g. MMSserver, operator specific platforms). GGSN exchanges

information about the mobile stationaAZs session statewith SGSN. This is mostly session establishment, modi-fication and teardown. In a nutshell, GGSN acts as pro-tocol translator between the aAIJmobileaAI world andthe external networks. On top of that, since the mobilestationaAZs connection is terminated on the GGSN, itserves as a connectivity anchor and assigns IP address tothe connected mobile station.

Connection type (or service type in other words) is de-fined in mobile networks by its APN (Access Point Name).APN determines what service does the mobile stationwant to access, what are the connection parameters (e.g.dynamic or static IP address allocation) etc. The con-nection/session in mobile networks language is denotedas PDP Context (Packet Data Protocol Context). It isbasically a virtual connection from the mobile station,through the whole mobile network to the service, iden-tified by various entities at different protocol levels (e.g.Tunnel End Point ID for the GPRS Tunneling Protocollevel, Temporary Logical Link Identifier on the Base Sta-tion Subsystem GPRS Protocol level, etc.). High levelarchitecture of GSM/GPRS network is depicted on Fig-ure 1.

Since the days of 2G data networks, network operatorshave deployed 3G (UMTS aAS Universal Mobile Telecom-

munications System) and its evolutions (HSPA aAS High

Speed Packet Access, HSPA+ aAS Evolved High SpeedPacket Access), 4G networks (LTE aAS Long Term Evo-

lution) and its evolutions (LTE-A aAS LTE Advanced,

LTE-A Pro aAS LTE Advanced Pro). Currently first 5Gnetworks are being deployed across the globe [17, 9].

3. Software Defined NetworkingSDN is in contrast with what is nowadays usually presentin production communication networks. The network gear(e.g. routers, switches or application gateways) combinethe network logic, management and data plane in one box.These boxes are usually managed though vendor specificcommand line interfaces (CLI) or by management appli-

Figure 2: OpenFlow packet processing pipeline[1].

cation of the given vendor.

SDN decouples control plane and user plane of the net-working gear and introduces standardized interfaces be-tween the two. By using feature set which is configurableby a well-defined protocol, it enables high vendor inter-operability and on top accelerates innovation, since dataplane and control plane can evolve each at its own pace(donaAZt need to evolve together as in case of traditionalnetwork gear).

The SDN approach used in the thesis is OpenFlow [1].It introduces simple model of packet processing. EachOpenFlow switch carries multiple OpenFlow tables (Fig-ure 2). Tables are filled with OpenFlow rules as in-structed by the SDN controller. The rules are composedof match field, against which the packet content is beingcompared (e.g. destination IP address value, UDP port,etc.), actions and instruction which shall be executed onthe packet if it is matched.

When a packet is received by the switch, it enters thefirst table and is being compared with the rules withinthe table. If the packet matches given rule, action set isbeing added to its metadata and finally being executed. Ifthe packet is not matched in any table, OpenFlow switcheither drops the packet or forwards the packet to the con-troller, where controller logic may decide what to do withit. Please note that OpenFlow switches do not carry anyof the logic that standard switch has (i.e. no MAC addresslearning or flooding mechanisms). OpenFlow switch sim-ply executes only the rules that are installed by the con-troller (and indeed in cooperation with the SDN controllercan mimic behavior of standard switches or routers).

In addition to match rules, actions and instructions, thereare also counters, which are incremented by rule hit andalso a possibility to use metering tables. OpenFlow fea-tures are being extended by every OpenFlow standardrelease and also allow to use so called experimenter ac-tions, which anybody can use while experimenting withnew, not yet standardized network approaches.

It needs to be noted, that there do exist many other SDNapproaches and OpenFlow is not the only one on the mar-ket, however it seems to be the frontrunner.

For example there is PCE (Path Computation Element)approach [24], which centralizes route computation andoptimization in the distributed environment of computernetworks. It basically offloads routers from CPU inten-sive task of route computation and optimization, whileusing also traffic engineering. Path computation clients(e.g. routers) communicate with the path computationelement by standardized Path Computation Element Pro-tocol (PCEP).

Information Sciences and Technologies Bulletin of the ACM Slovakia 3

Perhaps a competitor to OpenFlow in OpenStack basedinternal clouds, there is JuniperaAZs Contrail solution[16], which uses XMPP (eXtensive Messaging and Pres-

ence Protocol). In OpenStack environments, JuniperaAZsvRouter replaces vanilla OpenVSwitch and Contrail con-troller replaces standard Neutron (OpenStack default net-working) module. Contrail and vRouter communicate viaXMPP, while OpenVSwitch and Neutron though Open-Flow (and OVSDB aAS OpenVSwitch DataBase).

There are also SDN approaches, which did not gain com-mercial traction over the time (unlike OpenFlow). Forexample ForCES (Forwarding and Control Element Sep-aration) [4] is one of the oldest standards, which consistsof several RFC documents. Forces formally introduces thenotion of forwarding elements, control elements, forward-ing element managers and control element managers andtheir interfaces. Also, it decouples forwarding elements tological functional blocks. Due to its complexity, there ex-ist only few projects, which implemented ForCES and itwas finally overcome by much simpler, however not thatelaborative approaches (e.g. OpenFlow).

4. Related WorkResearch and development in the area of 3GPP mobilenetworks has been led by network infrastructure and chipsetvendors such as Ericsson, Nokia, Huawei, Qualcomm, Ciscoto name a few, partially in cooperation with mobile net-work operators.

This is due to the fact that mobile networks ecosystem, al-though with publicly available specifications is quite com-plex and up to now, there were almost no open-sourceimplementations of mobile nodes that would enable aca-demic community to execute practical evaluations of theirnovel and experimental approaches.

However nowadays (i.e. with advent of 5G networks), thesituation seems to be changing and both operators andnetwork equipment vendors are taking advantage of open-source projects and other innovative technologies rootedoutside of their usual ecosystem.

For example Huawei introduced a MobileFlow protocolbased architecture for 3GPP mobile networks. They in-vented this new protocol (althought most likely on thebasis of OpenFlow) and built architecture around it (Fig-ure 3), with MobileFlow forwarders similar to OpenFlowones, but considerably more feature rich. Supporting forexample also charging and GTP (GPRS Tunneling Pro-tocol) processing [22].

Ericsson on the other hand took a more conservative app-roach with extending standard OpenFlow protocol andOpenFlow forwarders with GTP support. Ericsson ar-gues that todayaAZs carrier networks rely on double rout-ing and forwarding decisions aAS one taken on the un-derlay layer (IP/MPLS routing), second on the overlay

layer (GTP routing). Centralized routing decision aASi.e. IP/MPLS and GTP logic centralized in a single pointshall simplify network design, optimize routing decisionsand accelerate failover [18].

In the area of academic research, some papers are stillbeing coauthored with the network gear vendors or ope-rators [10]. As for purely academic research, there is forexample an interesting paper from German researchers

Figure 3: HuaweiaAZs MobileFlow architecture[22].

that builds a theoretical model of the mobile network andevaluates benefits of SDN and NFV (Network FunctionsVirtualization) [19]. Other papers focus on the relatedtopics to SDN in mobile networks such as QoS [13] orits benefits to other technologies such as CDN (ContentDelivery Networks) [11].

Going forward the number of academic papers focusingon mobile networks is growing, however with focus on 5G,possibly 4G. There is little traction for the 3G networksand no traction at all in the area of 2G networks. De-spite that we think that applying SDN to 2G mobile net-work will be beneficial, as this network is still operationalworldwide with many terminals in network supportingthis technology exclusively. Although 2G, or GPRS tobe precise does not fulfill the requirements of todayaAZsinteractive multimedia applications (e.g. throughput, de-lay), it can be (and is) still used for example in context ofIoT (Internet of Things) for latency and throughput nondemanding use-cases, for example telemetry.

5. Thesis GoalsAs pointed out in the previous chapter, there are manyprojects aAS both from the industry and academia thatfocus on SDN in the area of 5G networks. Also LTE (4G)networks are being well used to provide use cases for SDNdeployment, which is visible also in related publications.We are not aware of any related work in regards to SDNin combination with GPRS (2G networks).

However in our opinion, second generation of mobile datanetworks is still relevant. According to the GSMAaAZsMobile Economy 2018 report and forecast, 40% of mobileconnections globally (out of 7.79 billion) were with 2Gtechnology. In developing regions, the number is evenhigher (e.g. Sub-Saharan Africa with 60% share of secondgeneration network) [14] [15].

It may be a fact that most of the future investments inregards to mobile networks will go towards either LTEextensions or new deployment of 5G. Thus making SDNresearch and development in the area od legacy networkseconomically not feasible for equipment vendors. HoweverGPRS network still remains in heavy use worldwide, ser-ving low speed and high latency data. GPRS ubiquity, its


low price and supported features can be and actually areused for data access where there is no other technology,or the connection features are good enough (e.g. alreadymentioned telemetry). Therefore our thesis will focus onSDN deployment in GPRS networks.

Thesis goals can be broken down into several incrementalsteps.

• Design and proposal of the new method for GPRScontrol and data / user plane separation. In orderto deploy software defined networking in GPRS ar-chitecture, control and user plane need to be sepa-rated. Unlike UMTS or LTE, in GPRS are theseinformation transported in a single stream of data.

• Design and proposal of new GPRS architecture basedon the SDN approach. Once there are available sep-arated streams of control and user plane data, soft-ware defined networking can utilized.

• Design and proposal of a simplified architecture fordelivery of GPRS services. Having SDN integratedin the mobile network, architecture will be simplifiedwith emphasis on the core network.

• Enhancement / generalization of the overall archi-tecture for other access technologies. With SDN con-troller in place and rest of the network having pro-grammable interfaces, architecture will be extendednot to provide GPRS only data service, but alsoto serve other access technologies while maintainingcommon transport core and control.

• Verification and evaluation of the new SDN basedarchitecture. After the concept is complete, one ac-cess technology will be selected and proof of conceptfor practical evaluation will be built. Also analyticalevaluation of selected feature of the new architecturewill take place.

6. Proposed ArchitectureNew SDN enabled architecture takes the existing stan-dard 3GPP GPRS as a baseline. Architectural changesare made with backwards compatibility in mind. Thisapplies to the new SDN enabled core compatibility withlegacy and standard radio access network. The rationalebehind this focus is to ease deployment of such solutionas much as possible. If any changes in the radio accessnetwork would be required due to the SDN architecturedeployment, it would be very hard to implement thesechanges cross whole radio access network, since it usu-ally consist of multiple hundreds or even thousands basestations which are geographically spread (depending onthe network operator footprint and also country terrainprofile).

New architecture is depicted on Figure 4. GGSN andSGSN were removed from the architecture and their func-tions were spread cross new nodes in the architectureaAS ePCU (enhanced Packet Control Unit), SDN con-troller, vGSN (virtual GPRS Support Node) and Open-Flow based forwarding core. Since there is no GGSN orSGSN anymore, also GTP (GPRS Tunneling Protocol) isnot used in the network. The only exception to GTP useis interfac-ing with legacy network operators for roamingpurposes. However, if interfaced with SDN enabled do-mains/operators only, no GTP is needed. Instead of GTP,

Figure 4: New SDN GPRS architecture.

Ethernet II header is being reused for tunneling purposesaAS we call this approach MAC tunneling.

6.1 Signaling and User Plane SeparationAs mentioned in previous section, in order to fulfil allsubsequent thesis goals, first we need to extract signalingand user data plane from the joint stream of data. A newnetwork element, which will execute this separation aASePCU was deployed on the interface between the radioaccess network and the core network - Gb.

Signaling messages on this interface are related eitherto SGSN-MS signaling (e.g. mobility or session mana-gement) or SGSN-BSC signaling (radio access networkmanagement). Separation of SGSN-BSC communication

is done on the GPRS-NS (GPRS aAS Network Service)protocol level, where all messages except NS-UNIDATAare being forwarded to the network element responsiblefor signaling handling (in our case this is vGSN, in stan-dard 3GPP architecture this is SGSN). MS-SGSN signal-

ing aAS data separation is being done on a higher levelaAS GPRS-LLC (GPRS aAS Logical Link Control) layer.LLC SAPI (Service Access Point Identifier) value deter-

mines the type of the payload aAS LL3, LL5, LL9 andLL11 being the user data related SAPIs and other valuesindicating signaling data.

6.2 New Nodes in the ArchitectureePCU is basically a special kind of OpenFlow forwarder,which understands GPRS protocols on the Gb interfaceand is able to identify and separate signaling from userplane data. Signaling is forwarded to another new nodeaAS vGSN, user plane traffic is handed over to OpenFlowtransport core.

vGSN processes only signaling messages (unlike SGSN),either mobile station signaling or BSC signaling. vGSNinterfaces also with SDN controller, which is responsiblefor whole core network, but also assists during authen-tication procedures or session management procedures.Actual session establishment and construction of the net-


Figure 5: Control plane in standard 3GPP GPRSarchitecture.

Figure 6: User plane in standard 3GPP GPRSarchitecture.

work path (towards Internet for example) is fully with theSDN controller.

SDN controller interfaces with the access network mana-gement and signaling node (vGSN), but also with the coretransport network (OpenFlow forwarders and ePCU) andsubscriber databases, in order to have subscription mana-gement logically centralized.

Transport core is based on OpenFlow compliant forwar-ders and is controlled by OpenFlow controller. These for-warders execute MAC tunneling according to OpenFlowrules set by the controller. Inner core forwarders manipu-late the Ethernet header only, however the access edgeforwarders (ePCU) examine the IP header and the accessspecific header (e.g. GPRS specific protocols). The exter-nal networks edge (e.g. Internet uplink) also examine theIP header in order to select correct tunnel for particularmobile station.

Architectural changes had naturally impact also on theprotocol stacks. Figure 5 depicts signaling plane in thestandard 3GPP GPRS network and Figure 6 depicts userplane. For the new GPRS SDN architecture, the protocolstacks are depicted on Figure 7 and Figure 8.

As mentioned before, standard 3GPP architecture emp-loys GTP, both for the signaling (GTP-C) and for userplane tunneling (GTP-U). In GPRS SDN architecture theGPRS specific signaling is terminated on vGSN, processed

Figure 7: Control plane in GPRS SDN architec-ture.

Figure 8: User plane in GPRS SDN architecture.

there and if controller cooperation is needed, controllerAPIs are being called via ReST (Representational statetransfer).

6.3 MAC TunnelingAs mentioned previously, GPRS SDN architecture doesnot use GTP as the core signaling and transport protocol.However, the need to separate users is still present.

At the beginning, we were considering to use somethingstandard and well known to the networking community(inventing new tunneling approach was not really a goalof thesis). However we found out that protocols such asGRE (Generic Routing Encapsulation), VxLAN (Virtualextensible Local Area Network), EoGRE (Ethernet overGeneric Routing Encapsulation), MPLSoGRE (Multipro-tocol Label Switching over Generic Routing Encapsula-tion) or even GTP, are poorly supported cross the open-source ecosystem of OpenFlow forwarders and controllers.Moreover we realized, that there is no need for a featurerich tunneling protocol, such as any of the mentionedones. Therefore we reused already present Ethernet IIheader for tunneling purposes (basically using the sourceMAC address as a tunnel ID. Having all traffic MAC tun-neled, one can steer and breakout traffic at any node thatcan match and set Ethernet II header fields (basically anyOpenFlow compliant forwarder).

6.4 Architecture Generalization for Other Access Tech-nologies

After SDN enabled GPRS architecture concept was fini-shed, we moved on with generalization of this architecturefor other access technologies. Following rules for applica-tion of the concept to other network types were proposed:

• Access specific protocols are terminated as close aspossible to the access network, on access adaptornodes.

• Access adaptor nodes extract access specific userplane data (ideally IP level) and forward it to trans-port core. If signaling is present, same node extractsit and forwards it to access network manager.

• Common transport core is based on MAC tunnels,which are enabled by OpenFlow forwarders and con-trolled by SDN controller.

• SDN controller is queried for session related proce-dures within access network, but also orchestratesprocedures across different access network managers.


Figure 9: Access agnostic SDN architecture basedon GPRS SDN architecture.

GPRS SDN architecture can be used for explanation ofthe evolution of the GPRS specific SDN concept to genera-lized SDN architecture (Figure 9). ePCU in GPRS SDNarchitecture is an access adaptor. Its responsibility is toterminate access specific protocols on the user plane(ifneeded, also to separate signaling from user plane data)and forward access specific signaling to access networkmanager aAS vGSN is the access network manager inGPRS SDN architecture.

Access network manager is responsible for access networkspecific signaling towards the client connected in particu-lar access network and for translation of requests to cont-roller (if request is SDN controller related aAS such assession establishment or authentication).

Controller is there to provide core network control, butalso orchestration cross multiple access managers and ac-cess adaptors in cases where client changes access technolo-gy. Controller also acts as a single authentication entity.

7. Architecture VerificationWe verified our concept using both theoretical and ex-perimental approach. In the theoretical verification, weexamined the new architecture efficiency in terms of userplane data transport. In experimental approach, we builta proof of concept of the architecture.

7.1 Theoretical VerificationIn theoretical evaluation part, we looked at the overheadof the MAC tunneling. It was compared to the populartunneling approaches used in the commercial carrier net-works and also to the GTP tunneling which it basicallyreplaced in the new architecture.

Thus as the baseline, we took the protocol stacks as de-fined by 3GPP (Figure 5 and Figure 6). GTP is usedhere as the tunneling protocol, however in real deploy-ment, it may be complemented with MPLS (MultiProto-col Label Switching) and perhaps also Ethernet VLANs(802.1q). As a sample packet distribution we used simple

Table 1: Simple IMIX DefinitionPacket size (IP Ratio Percentage Percentage oflevel) [Bytes] of packets traffic volume

[%] (at IP level) [%]40 B 7 58,3 % 6,8 %576B 4 33,3 % 56,4 %1500B 1 8.3 % 36,7 %

IMIX distribution of traffic (Table 1).

Since MAC tunneling does not need any additional pro-tocol headers, except Ethernet II which is in our caseanyways present, it proves itself to be the most efficientwhen compared to MPLS, 802.1q or even VxLAN.

7.2 Practical VerificationAs for practical evaluation, we have built a proof of con-cept of SDN powered GPRS network using real 2G BTShardware (i.e. no simulation or emulation used). This ac-cess technology was selected, due to the fact that at thetime, no other 3GPP access technology node was availablein the FIIT STU lab.

The setup was composed of Sysmocom SysmoBTS [23],which is a relatively inexpensive 2G (850/900/1800/1900MHz) BTS (Figure 10). It is designed, build and sold bythe community formed around Osmocom project. Thisproject started as a network security research, focusingmainly on 2G network security issues [20]. Over the time,the group developed own baseband software for few mo-dels of 2G phones and continued developing software andhardware of other 3GPP defined network nodes, (e.g. BSC,MSC, HRL, etc.) which they used during the network se-curity experiments.

SysmoBTS runs fully fledged ARM Linux, thus can hostthe whole mobile network and just connect its uplink tothe Internet (data) or to SIP PBX for voice (Session Ini-tiation Protocol Public Branch Exchange). Such setupis called network in the box, however we did not use itin this way. We ran only BTS and PCU applications onthe SysmoBTS, so the SysmoBTS acts just as standard(3GPP compliant) BTS. To the core network it exposes3GPP compliant Abis (voice/SMS) and Gb (data/SMS)interfaces. Both interfaces are logical and from the hard-ware point of view are terminated on the Ethernet portof the SysmoBTS unit. On the other side of the Ether-net cable we connected Linux PC running Ubuntu 14.04,64bit version.

Since our thesis is focused on mobile data, we will omitAbis (voice) going forward and focus on Gb. This in-terface is terminated on ePCU, which executes the dataplane aAS signaling separation function as described inthe previous section. User data is then forwarded to thetransport core, which is based on OpenFlow forwarders.Whole core, including the ePCU is based on experimentalopen-source OpenFlow forwarder implementation calledofsoftswitch13 [12]. Its code was modified by adding cus-

tom actions and match rules (for GPRS signaling aASuser plane separation), basically bringing GPRS aware-ness to the code. Ofsoftswitch13 is not designed to behigh performance software OpenFlow forwarder imple-mentation (unlike OpenVSwich), but rather focuses oneasy extensibility to provide a ground for network exper-


Figure 10: SysmoBTS hardware base station.

iments.

The egress part of core is basically aAIJInternetaAI up-link. Last forwarder in the MAC tunnel sets correct sourceand destination MAC addresses and sends packet to theLinux kernel, where it is matched with iptables rules andNAPT (Network Address and Port Translation) is ap-plied. Next the packet exits the Linux machine via sec-ond Ethernet interface (of Wi-Fi if applicable) towardsthe Internet.

Signaling data is forwarded from ePCU to the vGSN node,which is a combination of selected control functions ofSGSN and GGSN. vGSN processes communication withthe BTS and also with the mobile station. Osmo-sgsn [21]and openGGSN [6] code bases and surrounding librarieswere used to build vGSN. These open-sources eased theproof of concept development, mainly thanks to the imple-mented mobile protocol stacks and prototypes of GPRSspecific messages and state machines.

vGSN is on the north interfaced via ReST with the SDNcontroller. Controller holds full visibility of the networktopology, state of the forwarders, basically is in chargeof the whole transport network. Another open-sourceproject was used here. We built controller applicationon top of the Ryu framework [7]. Ryu provides hooks to

the OpenFlow processing aAS such as OpenFlow eventshandling, OpenFlow messages composition and parsing.However Ryu does not provide any controller, it is basi-cally a framework for building controllers. Thus all thecontroller logic is purely custom product of ours.

8. Conclusion and Future WorkWhile looking at the thesis goals, its contribution can bebroken down into following items:

• A new method of signaling-user data separation forthe GPRS network was proposed. This is based onthe SAPI information element and executed on theePCU node. According to our research, there doesnot exist similar approach for GPRS.

• Thanks to this separation, SDN was deployed in theGPRS network.

• GPRS architecture was simplified by removing SGSN,GGSN nodes and the GTP protocol, which was used

both for user data tunneling and also for signalingbetween the two nodes.

• New nodes were introduced to the architecture. Asmentioned before, ePCU is providing signaling-userdata separation, vGSN provides access network sig-naling processing, SDN controller is in charge oftransport core and connectivity orchestration. Open-Flow forwarding core is used as an underlay andprovides also overlay by using MAC tunneling.

• SDN enabled GPRS architecture was used as a basisfor proposal of an access agnostic (generic) SDN ar-chitecture. This architecture can integrate variousaccess technologies, while using the network conceptfrom GPRS SDN (such as common MAC tunnelsbased core controlled from SDN controller, accesscontrol by dedicated access managers aAS such asvGSN and signaling-user data separation providedby access adaptor (enhanced OpenFlow forwarders)

aAS such as ePCU.

• Architecture was verified by a proof of concept, us-ing unmodified BTS hardware and MS, which pro-ved that changes in the core network (implemen-tation of SDN concept) did not have impact on theradio access network and are transparent to it. Thiswill ease practical deployment of the concept. More-over effectivity of MAC tunneling was evaluated us-ing simple IMIX traffic model. Providing that MACtunneling does not use any additional protocol head-ers, it was the most efficient out of the comparedprotocols.

Future work on the topic can be done in various ways.One of them is definitely other access technologies inte-gration aAS for example LTE or WiFi.

Next there might be extension of the proposed GPRSSDN concept, where for example ciphering support canbe added. In the standard architecture, ciphering is exe-cuted on SGSN (for later network generations, this is be-ing pushed more to radio access network nodes). Sincethere is no SGSN, we removed this feature entirely fromthe concept. The reason is, that encryption, if neededis provided by the application layers. If implemented inGPRS SDN architecture, this would imply major changesmostly in the ePCU (and also in the controller). ePCUwould need to store ciphering keys and also temporaryidentities of the mobile stations (P-TMSI aAS PacketTemporary Mobile Station Identity). Moreover it wouldneed to examine signaling messages, in order to keep trackof P-TMSI reallocations and apply correct ciphering keyson the mobile stationaAZs traffic. In our opinion this goesfar beyond the simplicity of the forwarding plane (one ofthe concepts within SDN), thus was excluded from theoriginal proposal.

Also future work can be done in the area of MAC tun-neling, which as pointed out provides means of offloadingtraffic in any point of the network (as far as that point isa OpenFlow compliant forwarder).

Last, usual topic within computer networks research areawould be QoS (Quality of Service). MAC tunnels providea basis for different QoS model implementations. Diffe-rent traffic types, customers, access technologies can have


dedicated tunnels and these can traverse different for-warders and links with different properties (if multiplepaths are available between source and destination) andcan be optimized for given traffic type.

Acknowledgements. This project was partially sup-ported by the Tatra banka foundation under the contractNo. 2012et011.

References[1] Open networking foundation: Openflow switch specification

1.4.0, 2013.[2] 3GPP: 23.002: Network architecture; 13.1.0, 2014.[3] 3GPP: 23.060: General packet radio service (GPRS); service

description; stage 2; 13.1.0, 2014.[4] IETF: Forwarding and control element separation workgroup,

2015.[5] 3GPP: Specifications, 2019.[6] Open-ggsn, 2019.[7] Ryu SDN framework, 2019.[8] M. T. Abbas, T. A. Khan, A. Mahmood, J. J. D. Rivera, and

W. Song. Introducing network slice management insidem-cord-based-5g framework. In NOMS 2018 - 2018 IEEE/IFIPNetwork Operations and Management Symposium, pages 1–2,April 2018.

[9] J. Bannister, P. Mather, and S. Coope. Convergence Technologiesfor 3G Networks: IP, UMTS, EGPRS and ATM. John Wiley &Sons, Ltd, 2004.

[10] A. Basta, W. Kellerer, M. Hoffmann, K. Hoffmann, andE. Schmidt. A virtual SDN-enabled LTE EPC architecture: A casestudy for S-/P-gateways functions. In 2013 IEEE SDN for FutureNetworks and Services (SDN4FNS), pages 1–7, Nov 2013.

[11] J. Costa-Requena, M. Kimmerlin, J. Manner, and R. Kantola.SDN optimized caching in LTEen mobile networks. In 2014International Conference on Information and CommunicationTechnology Convergence (ICTC), pages 128–132, Oct 2014.

[12] CPqD. ofsoftswitch, 2019.[13] A. Elakkiya and P. Selvaraj. QoS based IP mobility management

scheme for the next generation SDN-LTE network. In 2018 2ndInternational Conference on Inventive Systems and Control(ICISC), pages 1355–1360, Jan 2018.

[14] GSMA. The mobile economy 2018. Technical report,https://www.gsma.com/mobileeconomy/wp-content/uploads/2018/05/The-Mobile-Economy-2018.pdf,2019.

[15] GSMA. Mobile economy 2018: Technology migration, 2019.[16] Juniper Networks. Contrail SDN, 2019.[17] H. Kaaranen, A. Ahtiainen, L. Laitinen, S. Naghian, and

V. Niemi. UMTS Networks: Architecture, Mobility and Services,Second Edition. John Wiley & Sons, Ltd, 2005.

[18] J. Kempf, B. Johansson, S. Pettersson, H. LÃijning, andT. Nilsson. Moving the mobile evolved packet core to the cloud.In 2012 IEEE 8th International Conference on Wireless andMobile Computing, Networking and Communications (WiMob),pages 784–791, Oct 2012.

[19] F. Metzger, C. Schwartz, and T. HoÃ§feld. GTP-based loadmodel and virtualization gain for a mobile network’s GGSN. In2014 IEEE Fifth International Conference on Communicationsand Electronics (ICCE), pages 206–211, July 2014.

[20] K. Nohl and C. Paget. GSM: SRSLY?, 2009.[21] osmocom. Osmo-sgsn, 2019.[22] K. Pentikousis, Y. Wang, and W. Hu. Mobileflow: Toward

software-defined mobile networks. IEEE CommunicationsMagazine, 51(7):44–53, July 2013.

[23] Sysmocom. SysmoBTS 1002, 2019.[24] J.-P. Vasseur and J.-M. L. Roux. RFC 5440 âAS path computation

element communication protocol, 2009.

Selected Papers by the Author

M. Nagy and M. Kotocová. An IP based security threat in mobilenetworks. In 2012 Proceedings of the 35th InternationalConvention MIPRO, pages 667âAS670, May 2012.

M. Nagy and I. Kotuliak. Enhancing security in mobile data networksthrough end user and core network cooperation. In Proceedingsof International Conference on Advances in Mobile Computing &Multimedia, MoMM âAZ13, pages 253:253âAS253:259, NewYork, NY, USA, 2013. ACM.

M. Nagy and I. Kotuliak. Utilizing openflow, SDN and NFV in GPRScore network. In V. C. Leung, M. Chen, J. Wan, and Y. Zhang,editors, Testbeds and Research Infrastructure: Development ofNetworks and Communities, pages 184âAS193, Cham, 2014.Springer International Publishing.

K. Burda, M. Nagy, and I. Kotuliak. Reducing keepalive traffic insoftware-defined mobile networks with port control protocol. InI. Khalil, E. Neuhold, A. M. Tjoa, L. D. Xu, and I. You, editors,Information and Communication Technology, pages 3âAS12,Cham, 2015. Springer International Publishing.

M. Nagy, I. Kotuliak, J. Skalný, M. Kalcok, and T. Hirjak. Integratingmobile openflow based network architecture with legacyinfrastructure. In I. Khalil, E. Neuhold, A. M. Tjoa, L. D. Xu, andI. You, editors, Information and Communication Technology,pages 40âAS49, Cham, 2015. Springer International Publishing.

R. Grežo and M. Nagy. Network traffic measurement andmanagement in software defined networks. In 2017 3rd IEEEInternational Conference on Computer and Communications(ICCC), pages 541âAS546, Dec 2017.

software deﬁned networking in wireless mobile...

Documents