SoftNet: A Software Defined Decentralized Mobile Network ...

WANG2_LAYOUT.qxp_Layout 1 3/13/15 1:57 PM Page 16

SoftNet: A Software Defined Decentralized Mobile Network Architecture toward 5G Hucheng Wang, Shanzhi Chen, Hui Xu, Ming Ai, and Yan Shi Abstract

Mobile cloud computing and other new emerging communication paradigms such as mobile social networking and Internet-of-Things, have brought challenges in flexibility, efficiency, and scalability to the current LTE network. Inspired by thinking of the fundamental mechanisms in LTE as reasons causing those problems, SoftNet, a software defined decentralized mobile network architecture toward 5G, is proposed in this article following principles proposed for designing an efficient and scalable network. The analysis of the working mechanisms of SoftNet, including its dynamically defined architecture, decentralized mobility management, distributed data forwarding, and multi-RATs coordination, show that SoftNet has improved system capacity and performance. Further, simulations are conducted to demonstrate that signaling cost in SoftNet, as an important performance metric, can be decreased significantly compared with LTE networks.

T

he development of smartphones and mobile communication technologies has led to many new network services, such as mobile social networking, mobile cloud computing, vehicular networking, and so on. These new services create diverse requirements for mobile networks. Moreover, according to comprehensively researched future communication scenarios, e.g. a total of 12 typical use cases studied by METIS, future communication paradigms have more stringent requirements for mobile networks, such as RTT (round-trip time) latency less than 10 ms for accessing cloud servers, the high speed of data transmission (at least 5 Gb/s for electronic offices), and extremely dense traffic (up to 900 Gb/s/km 2 for an open air festival) [1]. Such new requirements cannot be satisfied by current LTE networks, even in their evolved version, because of their inapplicable fundamental mechanisms, including centralized mobility management mechanism, centralized routing mechanism, heterogeneous wireless networks without coordination, and so on. Therefore, next generation mobile network architectures need to be redesigned to deal with the challenges of LTE networks. Recently, many organizations have proposed some good/ Hucheng Wang and Yan Shi are with the State Key Laboratory of Networking and Switching Technology at Beijing University of Posts and Telecommunications. Shanzhi Chen, Hui Xu, and Ming Ai are with the State Key Laboratory of Wireless Mobile Communications at China Academy of Telecommunications Technology. This work is supported by the National Natural Science Foundation of China for Distinguished Young Scholar (No.61425012), the Major National Science and Technology Special Project (No.2013ZX03001025-001), the National High-Technology Program (863) of China (No.2014AA01A701), and the China Next Generation Internet (CNGI) Project (No.CNGI-12-03-003).”

16

0890-8044/15/$25.00 © 2015 IEEE

innovative ideas for coping with challenges in current LTE networks. Nevertheless, most related work focuses on solving specific problems either in RAN or the core network by employing SDN (software defined networking) principles. For example, OpenRAN [2] proposes a software defined RAN (radio access network) architecture via virtualization. By virtualizing and allocating spectrum, computing, and storage resources to virtual access elements, the controller in the network can dynamically create and optimize virtual access elements based on service requirements. OpenRadio [3] decomposes wireless protocols into separate processing and decision planes, so that operators can just express decision plane rules and corresponding processing plane action graphs to assemble a data plane protocol running on a commodity multi-core hardware platform. Kempf et al. [4] enhances OpenFlow switches to support GTP (General Tunneling Protocol), so that the control plane EPC (evolved packet core) network function can be separated from specific network elements and implemented by software for openness and flexibility. The article also provides a detailed description of the extension of OpenFlow 1.2 protocol. Mobileflow [5] describes a concrete architecture for future carrier networks that builds on decoupling the control plane and the data plane, as well as a newly introduced MobileFlow stratum, thus the mobile carrier network employing SDN principles can support tunneling protocols such as GTP and PMIP. Softcell [6] realizes fine-grained and optimized policy control on base stations by supporting SDN switches, and it proposes data forwarding on core switches using hierarchical addresses and policy tags to minimize data plane state in the switches. In this article we propose what we refer to as a software defined decentralized mobile network architecture toward 5G (SoftNet), which is designed on the system level. SoftNet has the following advantages. First, by employing SDN and NFV (network function virtualization) technologies, SoftNet has excellent scalability and flexibility to accommodate different

IEEE Network • March/April 2015


communication scenarios. Second, signaling overhead in the core network is reduced dramatically due to decentralized mobility management. Third, both system capacity and performance are improved by supporting decentralized mobility management, distributed data forwarding, and multi-RATs coordination. Finally, the network control protocol can be simplified by adopting a RAT (radio access technology) independent control protocol. The remainder of this article is organized as follows. In the next section, the current LTE network architecture and challenges for the network are analyzed. After that, we first analyze principles for designing an efficient and scalable network, and then we describe the architecture of SoftNet following these principles. The working mechanisms of SoftNet are then introduced, after which we evaluate the signaling cost of SoftNet based on simulation results. We conclude the article in the final section.

Challenges for LTE Network Architecture At the beginning of designing LTE network architecture, the main service requirements come from IMS (IP multimedia subsystem) services and traditional Internet services. Howevber, with the advent of mobile Internet services and IoT (Internet of Things) services, many new service requirements are emerging for which LTE networks are not well suited. For example, as introduced in [7], mobile cloud computing requires seamless inter-system handover with low latency to guarantee the user’s experience, but in LTE networks, intersystem handover requires re-establishment the connection toward PDN GW (packet data network gateway), which increases handover latency and downgrades performance. 3GPP is working to enhance LTE networks to solve the issues created by new service requirements, and has completed several work items, including LIPA/SIPTO (local IP access/selected IP traffic offload) for offloading data traffic from EPC [8]; NIMTC/SIMTC (network improvement for machine type communication/system improvement for machine type communication) for supporting machine type communication in LTE networks [8]; and WLAN_NS (WLAN network selection) for optimizing WLAN selection for interworking [9]. However, fundamental mechanisms in LTE networks are still being challenged, e.g. the SIPTO mechanism introduced in 3GPP Rel 10 allows packet data network (PDN) connections toward SIPTO APN (access point name) to relocate their PGWs, but inefficient routing due to the centralized gateway still exists. Although NIMTC introduced an NAS (non-access stratum) level congestion control mechanism for suppressing signaling overload, it makes no enhancement to system capacity. WLAN NS allows optimized WLAN selection based on ANDSF (access network discovery and selection function) policies, whereas independent radio access technologies cannot co-operate for optimal radio resource utilization. In the following sections, we summarize the challenges for current LTE networks that hinder LTE network from efficiently meeting the new service requirements.

Limited System Capacity New emerging communication paradigms such as mobile cloud computing require much more radio resources for high speed data transmission. However, radio resource utilization in current LTE networks is inefficient, because neither the unified schedule of radio resources of different RATs nor coordination among different RATs is supported. In addition, at least one PDN connection needs to be maintained in the core network for keeping each UE (user equip-


ment) “attached,” which is useful to IMS voice service for callback and quickly restoring end-to-end connection. However, this mechanism may not be suitable for some IoT services, such as smart metering or monitoring services which usually need to be supported by massive number of devices, because lots of network resources will be consumed by a huge number of PDN connections established for infrequent data transmission, and ultimately system capacity will be decreased.

High Signaling Overhead Most mobile Internet applications in smartphones support push services over IP. Push services, such as Apple Push Notification, require a persistent TCP connection between smartphone and notification server to transfer a very small packet [10]. Therefore, the LTE network employing a centralized mobility management mechanism has to spend lots of signaling to maintain PDN connections over which the TCP connections are kept. In addition, some mobile Internet services or IoT services handle small data transmissions, e.g. OTT (over the top) services that can cause high signaling overhead in the LTE network to maintain a data forwarding path. Moreover, a large number of MTC devices trying to access the network simultaneously may cause signaling overload in the core network because of a centralized mobility management mechanism, e.g. monitoring devices deployed for environment monitoring or data collection may try to access a mobile network simultaneously for reporting detected events.

Inefficient Data Forwarding In LTE networks, all data traffic has to be routed via a PGW to enter IP networks. As all user equipment within an operator’s network is served by a small number of centralized PGWs, the serving PGW is not changed even if a UE moves a long distance, which makes data forwarding along an established PDN connection very inefficient. For example, a UE established a PDN connection toward a PGW in place A for accessing an application server in place B. Some time later, even if the UE moves to place B, the serving PGW is still the PGW located in place A. So the entire operator’s network becomes a private IP sub-network. In addition, centralized data forwarding requires that data traffic always traverses EPC even if some data traffic is going to/from local application servers, e.g. enterprise cloud servers. Therefore, a centralized routing mechanism in an LTE network has defects of long latency, data forwarding inefficiency, and user plane congestion. Furthermore, a data forwarding path in an LTE network consists of multiple tunnels, with the end points of each tunnel supporting tunneling protocols, e.g. the GTP protocol, while the intermediate nodes along each tunnel are transparent to the tunneling protocol. The routing mechanism among these intermediate nodes is still based on traditional IP routing protocols, consequently, the data forwarding path cannot be optimized by considering network conditions and service requirements completely.

High Cost and Poor Scalability In LTE networks, all network elements must be specialized, especially those supporting both control and data planes, e.g. SGW (serving gateway) and PGW, and the interaction between any network elements, must be based on a standardized interface. Therefore, the openness and flexibility of the network are extremely limited, with the consequences being that operators’ capital expenditures for scaling up the network are high, and the operator’s innovations on existing network elements, such as developing new network services, are very difficult.

17


Policy control

Communication control

Network ... management

Network controller

Virtual network

WLAN AP

Virtual compute Virtual storage

Operator’s service networks SDN based core network

BTS/NodeB/eNB

Cloud computing

CDN server

5G base station

Internet

Multi-RATs Decentralized Gateway coordination control control NFVI

Figure 1. Architecture of SoftNet.

A Software Defined Decentralized Mobile Network Architecture Toward 5G (SoftNet) According to the analysis in the previous section of the challenges for LTE networks created by new communication paradigms, we found that some fundamental mechanisms of LTE networks become inapplicable to serve new emerged services any longer. Therefore, we propose to re-design the mobile network architecture instead of enhancing existing LTE networks to cope with those challenges. In this section, we first put forward the principles for designing an efficient and scalable mobile network, and then we propose a new network architecture called “SoftNet” that complies with the proposed design principles.

proof network architecture, since it not only means flexibility and low cost to support unexpected services and scale up the network, but also implies that new services can be deployed quickly and new market demands can be fulfilled in time. Simplicity: Current cellular mobile networks employ lots of protocols involving multiple network elements to perform mobility management and/or establish data forwarding tunnels, e.g. establishing a GTP tunnel involves an MME (mobility management entity), eNB (evolved NodeB), SGW, and PGW. Mobility management in current 3G and LTE networks needs to be supported by different NAS protocols. It is doubtful that so many protocols have to be implemented in future mobile networks as they incur complexity, inefficiency, even function redundancy of communication system [11]. Therefore, simplifying protocols should be one of objectives of designing new network architectures.

Design Principles of SoftNet

The Architecture of SoftNet

In order to design a mobile network architecture with high performance to overcome the challenges analyzed in the previous section and satisfy new service requirements, we propose the following principles based on previous work on designing cellular mobile networks, investigation of new service requirements, and research on new network technologies. Adaptability: There are different communication paradigms with distinct characteristics in future mobile networks, so it is impossible for the network to adopt the same approach to efficiently serve diverse types of services. Therefore, future mobile network architectures must be adaptable to cope with different communication scenarios. Efficiency: The analysis of LTE network shows that new mobile communication paradigms may cause high signaling overhead, inefficient data forwarding, and/or long RTT latency in LTE networks, which ultimately leads to inefficient network resource utilization. Therefore, supporting new communication paradigms with efficient network resource utilization becomes a vital pursuit of re-designed mobile network. Scalability: Scalability is important for designing future-

With the guidance of the design principles discussed above, SoftNet, a novel architecture for future mobile networks is proposed as shown in Fig. 1. SoftNet adopts decentralized network control on the system level to improve flexibility and scalability, but chooses centralized control on the component level to promise the efficiency of network resource utilization. SoftNet consists of a unified radio access network and an SDN based core network. In SoftNet, all radio access points in unified RAN must be connected with access servers that are at the edge of an SDN based core network, so that mobile terminals served by the radio access points can either visit the operator’s service networks or third party service platforms, such as a cloud computing platform, via the core network, or access the Internet or CDN (content delivery network) server via a distributed gateway function within the access server. The control plane network functions are supported in the SDN based core network and unified RAN separately. The network functions supported in the SDN based core network mainly include communication control functions (CCF)

18



Multi-RATs coordination

Distributed control

Gateway control

Internet/

CDN responsible for mobility management, policy control function supporting QoS (Quality of Service) and network policy control, and network management function (NMF) monitoring Virtual Virtual Virtual Operator’s network conditions and defining network archistorage computing switch service SDN based core tecture. Network functions in the unified RAN networks network are deployed on an access server, which at least includes a multi-RATs coordination function, Traffic offloaded decentralized control function (DCF), and gateTraffic routed via core network way control function. The multi-RATs coordination function can monitor wireless network Figure 2. Traffic flows in SoftNet including traffic offload at access server and conditions and steer user traffic among selected traffic routed via SDN based core network. RATs. The DCF is responsible for decentralized mobility management that allows mobility events to be handled by decentralized control be established. The access server also relies on NFV architecplane network elements. The gateway function allows mobile ture to improved system scalability. terminals to access the Internet or CDN without traversing DCF is another basic component in SoftNet to support the core network. The data plane network elements in SoftNet decentralized mobility management. While mobile terminals are implemented on SDN switches, including a distributed move from a radio access point to another served by the same gateway in the RAN and egress gateways in the core network. access server, any location management or handover management signaling is handled by DCF. But if mobile terminals SDN Based Core Network — The core network of SoftNet is move between radio access points served by different access designed as an SDN flavor network consisting of network servers, DCFs in different access servers have to cooperate functions, network controllers, and network infrastructures. with CCF to perform location management and handover The network functions in the SDN based core network mainly management. The DCF is also responsible for paging hantake responsibility for centralized network control such as dling when it is triggered by signaling from CCF or an arriving admission control, QoS control, network management, and so downlink data packet. on. The network controller consists of an SDN controller and The multi-RATs coordination function in the access server virtual network function (VNF) orchestrator. The network allows the network to select RATs and steer user’s traffic infrastructures in the core network include NFV infrastrucamong selected RATs. Further, by virtualizing radio resources tures (NFVI) and physical equipment. of wireless networks, this function can even act as a controller A basic network function in the core network is CCF. for scheduling virtualized radio resources, as introduced in [2]. The CCF can be instantiated for different usages based on In order to support efficient data forwarding, the distributthe defined network architecture. If SoftNet is defined as a ed gateway function is supported in the access server for trafdecentralized network by NMF, the CCF will be instantiatfic offloading. The control plane of the distributed gateway is ed to manage mobility events out of the access server conimplemented as a gateway control function in the access servtrol. Otherwise, the CCF will be instantiated to manage all er by virtue of NFV. When mobile terminals would like to mobility events. Another important network function is the establish connections for accessing the Internet, local netpolicy control function which not only provides policies for works or CDN, the gateway in the access server, will act as a network management, but also determines the parameters mobility anchor for data forwarding, and allocate an IP for QoS control dynamically. The policy control function address for each connection. For data traffic of services for may connect with cloud computing services providing big which traffic offload is not activated, such as IMS voice calls, data analysis on user behaviors/habits in order to generate the access server has to forward the data traffic to the core appropriate policy parameters. The NMF in the core netnetwork. The traffic flows in SoftNet are shown in Fig. 2. work is used to determine network architecture and manage all network functions. Based on received instructions from NMF, the VNF orchestrator in the network controller can configure virtual machines Analysis of the Key Working Mechanisms of (VMs) to corresponding VNFs, and/or calculate a VNF forSoftNet warding graph. The VNF forwarding graph has to be sent to the SDN controller to generate corresponding flow rules. To support new features of decentralized mobility manageFinally, the generated rules will be installed on related virtual ment, distributed data forwarding, multi-RATs coordination, switches and/or physical switches. and RAT independent network control, SoftNet works in different ways to improve system capacity and performance. Unified RAN — MDN [11] points out that the deployment of Dynamically Defined Architecture mobility anchors, centralized or distributed, is critical for mobility management efficiency. In SoftNet, access servers SoftNet has the adaptive ability to accommodate different with gateway functions are deployed as distributed mobility communication scenarios. According to the network configuanchors in unified RAN to support decentralized mobility ration, the operator’s policies, and/or collected information management. Moreover, for a unified schedule of radio including the number of users in a particular area, data traffic resources of different wireless networks, the multi-RATs coordensity, user’s mobility information, status of network eledination function is supported in the access server as well. ments, and so on, NMF can determine a corresponding netTherefore, each access server becomes a neuron of SoftNet. work architecture by enabling/disabling related virtualized Radio access points, including future base stations such as 5G network functions for efficient network resource utilization. base stations, and existing base stations including eNB, For example, in a business district, the NMF can enable DCF NodeB, BTS, WLAN AP, and so on, can be served by access to perform mobility management locally during the day to servers as long as the connections with the access server can reduce signaling overhead in the core network, but disable it


19


User’s preference

Authorized QoS parameters

Operator’s policies

networks, which can be easily estimated, for example, for paging a UE with a TA (tracking area) list coving 20 eNBs, MME has to send paging requests to all 20 eNBs Monitored and all eNBs have to page the UE. However, in SoftNet, Historical Mobility Logic for wireless location assuming each access server is serving five eNBs, only information determining RAT networks information nine paging request messages are needed (four messages conditions from the CCF to the access server, and five messages from the access server to eNBs), and only five eNBs have to page the UE. Gathering information of Decentralized handover management allows intraSelected RAT each RAT access server handover to be handled by the access server, and only inter-access server handover has to be handled by the CCF. Intra-access server handover has no Figure 3. Logic for selecting RAT based on QoS requirement, wiresignaling to the core network, thus signaling overhead is less network conditions, user’s mobility information etc. obviously less than that of X2 handover in LTE networks. Even though signaling overhead of inter-access server handover is slightly more than that of X2 handover at night to save power. In addition, the NMF can allocate/deprocedure, the overall signaling overhead for handover manallocate network resources to network functions based on agement can still be decreased compared with LTE networks monitored network conditions, so that the network resources by optimizing network deployment. can be distributed with optimal proportion.

Enhancing System Capacity

Improving Efficiency of Data Forwarding

With the support of multi-RATs coordination in unified RAN, the access server can monitor the conditions of each RAT, and then select the most suitable RAT(s) to establish radio connection(s) and/or steer user traffic among the established connections by taking service QoS requirements, user’s preference, user’s mobility information, monitored condition of each RAT, and/or the operator’s policy/configuration into account as shown in Fig. 3. The efficient utilization of radio resources in the overall system can ultimately lead to enhancement of RAN capacity. Distributed data forwarding can also enhance the entire network capacity significantly. First, a distributed gateway function in the unified RAN can improve network throughput of data traffic going to/from the Internet, local networks, or CDN. Second, network resources in the core network can be saved to serve more mobile terminals because of traffic offloading in RAN. Finally, since most Internet streams have very short lifetimes [12], connections established for traffic offloading can be disconnected and/or re-established on demand, which also enhances the system capacity. Finally, system capacity can be enhanced by decentralized mobility management which can decrease signaling overhead in the core network. In SoftNet, decentralized mobility management allows the access server to handle mobility events locally, e.g. intra-access server handover only requires the access server to switch the data forwarding path internally, so that less mobility management signaling needs to be handled by the network than in LTE networks, which potentially enhances system capacity on signaling handling.

In SoftNet, efficiency of data forwarding can be improved significantly with the help of distributed data forwarding, which is supported by the distributed gateway function deployed in the unified RAN. When the DCF in the access server receives authorization to handle a connectivity request from a mobile terminal locally, the DCF will establish a connection between the mobile terminal and the gateway function, so that data traffic can be forwarded directly via the gateway function. Some applications with small data transmission in SoftNet may send only a few bytes each time. In this case, there is no need to establish dedicated connections for those applications. Rather, the small data packet can be piggybacked onto a signaling message and parsed by the access server; then the data packet can be routed via the distributed gateway directly, or transferred along a shared data path from the access server to the egress gateway.

Reducing Signaling Overhead Since research on human mobility patterns reveal that humans have a strong tendency to return to locations they visited, and the probability distribution of their gyration radius follows truncated power-law [13], we can infer that if access servers are deployed in areas with high frequent human activities, such as business districts and residential areas, most mobility events will be handled by access servers, and signaling overhead in the core network will be decreased dramatically. With the help of decentralized location management in SoftNet, location update signaling sent to the CCF in the core network can be decreased, because movement of a mobile terminal is not visible to the CCF as long as the mobile terminal is still served by the same access server. Moreover, paging handling has much less signaling overhead than that in LTE

20

Ensuring Flexibility and Scalability with Lower Cost Scalability of SoftNet can be enhanced by employing SDN and NFV technologies. Since the control plane and the data plane are separated in the SDN based core network, the controller can exploit complete knowledge of the network to optimize flow management and support service-user requirements of scalability and flexibility [14]. On the other hand, all forwarding equipment can be installed and managed flexibly via the standardized southbound interface. Additionally, the network controller provides an open northbound interface to ensure that any new network functions can be realized and deployed rapidly, which encourages network innovations significantly. NFV can decouple network functions and hardware resources by virtue of virtualization [15], thus any network functions in SoftNet can be implemented by software, such as DCF in the access server and CCF in the SDN based core network, and any hardware resources including those for general purpose can be virtualized to VMs.

RAT Independent NAS Protocol In order to simplify the NAS protocol, SoftNet can employ network control as a system service. Therefore, the network control protocol entity on the mobile terminal side can be implemented as a system component of the mobile OS, and network services including mobility management and connection management can be provided to mobile terminals accessing the network via any RAT. In addition, since a software component can be upgraded easily, the network service can be



SoftNet

LTE network

Networking

Dynamically defined by software

EPC is statically configured via OAM

Risk of signaling overload

Low decentralized mobility management reduces signaling overhead in core network

High centralized mobility management increases the risk of signaling overload

Efficiency of data forwarding

High flattened data plane terminates at distributed gateway improves the efficiency of data forwarding

Low data forwarding tunnels need to be established involving eNB, SGW and PGW

Support of heterogeneous networks

Good multi-RATs are coordinated under control of access server

Poor multi-RATs are less coordinated in RAN

Efficiency of network resource utilization

High network resources are centralized scheduled with global view of network

Low network resources are scheduled with an autonomous system view

Scalability of network

Excellent commodity hardware can be virtualized for different usages

Poor specialized hardware is required

Portability/scalability of network function

Excellent network functions are implemented by software and running on VMs

Poor network functions are tightly coupled with physical equipments

Robustness

High robustness is enhanced by decentralized mobility management, distributed data forwarding and multi-RATs coordination

Low single node of failure caused by centralized mobility management and centralized data forwarding

NAS protocol

RAT independent protocol employed as system service provides flexibility and scalability to upgrade network services

Difficult to be upgraded

Table 1. Comparison between LTE network and SoftNet. upgraded flexibly to support new features. In summary, compared with LTE networks, SoftNet can significantly improve system efficiency in networking, mobility management, data forwarding, radio resource management, and so on. Table 1 shows the differences.

Performance Evaluation In this section, the performance of SoftNet is evaluated based on a comparison of signaling cost with LTE networks. Signaling cost here is given by the product of size of the mobility management message and the weighted distance (hops). To simplify the simulation, we make the following assumptions: • The mobile terminal is always in connected mode and passes through the networks back and forth by way of uniform rectilinear motion. • Control messages are the same size of a unit length. • The distance between the network controller and the SDN switch is set to 1 hop by assuming a dedicated line is used for the OpenFlow control channel. • In LTE networks, handover between eNBs served by the same MME is X2 handover; handover between eNBs served by different MMEs is S1 handover. • During inter-access server handover, the network control establishes direct forwarding paths between the source access server and the target access server for both the control plane and the data plane. Our simulations tested the signaling cost in LTE networks and in SoftNet separately with parameters given in Table 2. The simulation scenarios are as follows: 18 eNBs are deployed in a line with interval of 500 m; each access server serves three eNBs in SoftNet, and each MME serves nine eNBs in LTE networks. In the LTE network, we assume that the eNBs, which could be served by the same access server in SoftNet, have the same distance to the MME. The simulation results shown in Fig. 4 demonstrate that signaling cost in SoftNet on average is 50 percent less than in LTE


SoftNet

LTE network

Distance between eNB and access server (hops)

1

N/A

Distance between eNB and MME (hops)

N/A

15

Distance between access server and CCF (hops)

15

N/A

Distance between MMEs (hops)

N/A

15

1

N/A

500 m

500 m

Parameters

Distance between access servers (hops) Radius of eNB coverage Velocity of mobile terminal

2 m/s

Table 2. Parameters for simulations.

networks. In the simulation, when the mobile terminal moves across the boundary of an area served by an access server, it will perform inter-access server handover, which will cause more signaling cost than in an X2 handover. Therefore, to optimize the network access servers should be deployed to cover areas where plenty of mobile terminals remain a long time. As a result, most handovers are intra-access server handovers.

Conclusion In this article, based on analysis on the challenges for LTE networks brought about by new communication paradigms, SoftNet, a new network architecture consisting of an SDN based core network and a unified RAN, is proposed to deal

21


2000 1800

SoftNet LTE

1600 Signaling cost

1400 1200 1000 800 600 400 200 0 20

40

60

80

100

120

140

160

180

200

Time (*20s)

Figure 4. Variation of signaling cost over time. with these challenges. SoftNet is a flexible and scalable system that can dynamically enable/disable related virtual network functions and employ new working mechanisms to improve the efficiency of network resource utilization. Thus SoftNet improves system capacity and performance. The performance evaluation is based on signaling cost, and the simulation result shows that SoftNet, by employing decentralized mobility management, has decreased signaling cost compared with LTE networks. Therefore, we believe future mobile networks are evolving to “decentralize,” whereas issues such as unified radio resource management, QoS mapping on radio links, and coordination between access servers still need further study.

Acknowledgment We thank Dr. Heli Zhang of Beijing University of Posts and Telecommunications for her comments and suggestions, which improved the overall quality of this article.

References [1] P. Popovski et al., ICT-317669-METIS/D1. 1 Scenarios, Requirements and KPIs for 5G Mobile and Wireless System, Tech. Rep., 2013; available: https://www. metis2020. com/ [2] M. Yang et al., “OpenRAN: A Software-Defined RAN Architecture via Virtualization,” Proc. ACM SIGCOMM, Aug. 2013, pp. 549–50. [3] M. Bansal et al ., “Openradio: A Programmable Wireless Dataplane,” ACM Proc. First Workshop on Hot Topics in Software Defined Networks, Aug. 2012, pp. 109–14. [4] J. Kempf, et al., “Moving the Mobile Evolved Packet Core to the Cloud,” IEEE Wireless and Mobile Computing, Networking and Communications (WiMob), Oct. 2012, pp. 784–91. [5] K. Pentikousis, et al., “Mobileflow: Toward Software-Defined Mobile Networks,” IEEE Commun. Mag., vol. 51, no. 7, Jul. 2013, pp. 44–53. [6] X. Jin et al., “Softcell: Scalable and Flexible Cellular Core Network Architecture,” Proc. Ninth ACM Conf. Emerging Networking Experiments and Technologies, Dec. 2013, pp. 163–74. [7] Z. Sanaei et al., “Heterogeneity in Mobile Cloud Computing: Taxonomy and Open Challenges,” IEEE Commun. Surveys & Tutorials, vol. 16, no. 1, 2014, pp. 369–92.

22

[8] “General Packet Radio Service (GPRS) Enhancements for Evolved Universal Terrestrial Radio Access Network (E-UTRAN) access,” Release 12, 3GPP TS 23.401, 2013. [9] “Architecture Enhancements for Non-3GPP Accesses,” Release 12, 3GPP TS 23.402, 2013. [10] B.-J. J. Kim et al., “Directions for Future Cellular Mobile Network Architecture,” First Monday, vol. 17, no. 12, Dec. 2012. [11] S. Chen et al ., “Mobility-Driven Networks (MDN): From Evolution to Visions of Mobility Management,” IEEE Network , vol. 28, no. 4, July/August 2014, pp. 66–73. [12] N. Brownlee et al., “Understanding Internet Traffic Streams: Dragonflies and Tortoises,” IEEE Commun. Mag ., vol. 40, no. 10, Oct. 2002, pp. 110–17. [13] M. C. Gonzalez et al., “Understanding Individual Human Mobility Patterns,” Nature, vol. 453, Jun. 2008, pp. 779–82. [14] S. Sezer et al., “Are We Ready for SDN? Implementation Challenges for Software-Defined Networks,” IEEE Commun. Mag. , vol. 51, no. 7, Jul. 2013. [15] “Network Functions Virtualisation; End-to-End Architecture,” Draft ETSI GS NFV-0010 V0.1.4, 2013.

Biographies HUCHENG WANG ([email protected]) received his M.S. degree from Beijing University of Posts and Telecommunications (BUPT) in 2008, and now is working as a doctoral student in Communication and Information System at the State Key Laboratory of Networking and Switching Technology, BUPT. He is also a senior standards engineer at the China Academy of Telecommunication Technology (CATT). His research interests include architectures, networking, and protocols for cellular mobile network. SHANZHI CHEN [SM’04] ([email protected]) received his Ph.D. degree from Beijing University of Posts and Telecommunications (BUPT), China, in 1997. He joined Datang Telecom Technology & Industry Group in 1994, and has been serving as CTO since 2008. He was a member of the steering expert group on information technology of the 863 Program of China from1999 to 2011. He is a member of Advisory Committee of Experts on the IoT development of China, the director of State Key Laboratory of Wireless Mobile Communications (CATT), and a board member of Semiconductor Manufacturing International Corporation (SMIC). He has made outstanding contributions to the development from TD-SCDMA 3G to TD-LTE-advanced 4G. He received the State Science and Technology Progress Award of China in 2001 and 2012. His current research interests include network architecture, wireless mobile communication, IoT, and emergency communication. Hui XU ([email protected]) received her Ph.D. degree from Xian Jiaotong University in 1999. She is now the manager of the Ubiquitous Network Department at Datang Wireless Mobile Innovation Center, and is involved in research of key technologies in Internet Of Things (IoT) and machine to machine communications (M2M). M ING A I ([email protected]) joined Datang Telecom Technology & Industry Group in 1998. Since 2008 he has been participating in 3GPP CT1 and SA2 meeting as a standards delegate and a coordinator of Datang. Before 2008, he worked as a software engineer and R&D manager for telecommunication equipment. His research interests include mobile communication technology, Internet technologies, and standards activities. YAN SHI ([email protected]) received her Ph.D. degree from Beijing University of Posts and Telecommunications (BUPT) in 2007. She is currently a member of the research staff of the State Key Laboratory of Networking and Switching Technology, BUPT. Her current research interests include network architecture evolution, protocol design and performance optimization of future networks and mobile computing, especially mobility management technology.
