DM3: distributed mobility management in MPLSbased ...

INTERNATIONAL JOURNAL OF NETWORK MANAGEMENT Int. J. Network Mgmt 2014; 24: 85–100 Published online 22 December 2013 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/nem.1854

DM3: distributed mobility management in MPLS-based access networks Javier Carmona-Murillo,1,*† José-Luis González-Sánchez,2 David Cortés-Polo2 and Francisco-Javier Rodríguez-Pérez1 1

Department of Computing and Telematics System Engineering, University of Extremadura, Cáceres, Spain 2 Research, Technological Innovation and Supercomputing Center of Extremadura Cénits, Cáceres, Spain

SUMMARY Over the last few years, mobility management in the Internet has been one of the most active fields in communications. The recent increasing mobile traffic demand is having an important impact on the design of mobile networks and some limitations are arising from traditional network deployments. In order to deal with this new scenario, mobility management network architectures are being redesigned towards a more distributed operation. In this paper, we introduce DM3 (distributed mobility management MPLS), a fully distributed architecture designed to track efficiently the mobility of users in the current paradigm of evolving mobile IP networks. In DM3 architecture, several nodes are distributed in the MPLS-based access network and the mobile nodes are served by a close-by mobility anchor. With this operation, we reduce the routing and registration update costs, and provide a low handoff latency with a minimal packet loss rate. Analytical and experimental results are presented to justify the benefits of our proposed architecture. Copyright © 2013 John Wiley & Sons, Ltd.

Received 20 May 2013; Revised 18 September 2013; Accepted 21 November 2013

1. INTRODUCTION Mobile data traffic in the Internet has experienced an exponential growth over the last few years due to the development and deployment of multiple wireless networks. Devices with different network interfaces have become commonplace and users expect to be always connected from anywhere at anytime. In this heterogeneous scene, moving towards all-IP architectures, the IETF (Internet Engineering Task Force) has standardized a fair number of IP mobility management protocols such as Mobile IPv6 [1] or Proxy Mobile IPv6 [2]. These protocols, host-based and network-based respectively, are widely accepted as the most appropriate mechanisms for addressing IP mobility management in future wireless networks. Moreover, new challenges arise in the evolution of mobility management in the Internet: firstly, the need to solve the limitations of the traditional structure of cellular networks, where a small number of centralized anchors manage the traffic of millions of mobile nodes [3]; secondly, the ability to provide a similar level of quality of service (QoS) while a user moves among heterogeneous networks [4,5]. Regarding the first challenge, current mobility management schemes developed for IP and cellular networks rely on a centralized mobility anchor entity. This node is responsible for both mobility signaling and user data forwarding. This centralized approach is likely to have several issues or limitations, which require costly network dimensioningand engineering to resolve. The main problems identified concerning centralized solutions are: non-optimal routing, scalability issues and excessive signaling overhead, which implies longer handover latencies and vulnerabilities due to the existence of a single point of failure [6]. *Correspondence to: Javier Carmona-Murillo, Department of Computing and Telematics System Engineering, Polytechnic School, Av. Universidad s/n, 10003, Cáceres, Spain. † E-mail: [email protected] Copyright © 2013 John Wiley & Sons, Ltd.

86

J. CARMONA-MURILLO ET AL.

The second goal, related to assuring the provision of sufficient network resources, has been largely studied in both wired and wireless environments. MPLS natively supports tunneling and also offers fast forwarding times. In addition, MPLS with its traffic engineering (TE) is a QoS technology introduced to enhance the performance of the Internet’s datagram model in terms of both management and delivery. Its integration with Mobile IP has worked successfully due to the ability of MPLS to engineer traffic tunnels efficiently, including constraint-based routing, survivability and recovery, as well as avoiding congestion and enabling an efficient use of the available bandwidth [7,8]. These features make MPLS a potential technique to solve Mobile IP’s operational and architectural shortcomings, such as high handoff latency and packet loss or high global signaling load and scalability issues [9]. MPLS can also be viewed as a tunneling technology that overcomes the tunneling techniques proposed in Mobile IP standard [10] owing to its reduced operational overhead (the 40-byte IPv6 tunnel header is replaced by a 4-bytes MPLS label [11]. The major contributions of this paper are threefold: (i) we present a fully distributed mobility architecture called DM3 (distributed mobility management MPLS), which addresses the limitations of centralized mobility architectures and leverage on the distributed mobility management (DMM) paradigm; (ii) an analytical model has been developed to compare our proposal with other well-known approaches. The results show that the DM3 operation can reduce routing and registration update costs, provide low handoff latency and minimize the packet loss rate during movements; (iii) an experimental testbed has been built in order to evaluate the impact of the location of the distributed mobility anchor in the performance of the architecture. Moreover, to thebest of our knowledge, this article is the first to present qualitative and quantitative analysis on distributed mobility management solutions and MPLS integration. The rest of the paper is organized as follows. In Section 2, we present background knowledge about centralized and distributed mobility management protocols. In Section 3 our DM3 proposal is presented in detail. We have developed an analytical model used to compare our solution with other similar solutions in the literature. This is presented in Section 4. In Section 5 the performance evaluation and experimental results are shown. Finally, concluding remarks are given in Section 6.

2. RELATED WORK In this section we present a brief overview of the centralized mobility management protocols used in the current mobile networks, discuss its integration with MPLS and introduce the DMM paradigm. A brief description of the solutions, whichare compared with our DM3 proposal, is also given.

2.1. Centralized IP mobility management Mobility management in the Internet is a key aspect of mobile communications and is the next step in the Internet evolution. It is practical now for a mobile node to roam between different access technologies and furthermore it is reasonable to expect address continuity and session persistence across these handoffs. With these requirements, the IETF has developed Mobile IPv6 and Proxy Mobile IPv6. Mobile IPv6 is a host-based protocol requiring the participation of the host in all aspects of mobility management, whereas Proxy Mobile IPv6 is a network-based approach where the host does not participate in any mobility-related signaling. In this paper, we focus on Mobile-IP solutions. A further explanation of IP-based mobility management protocols can be found in Kong et al. [12]. In order to enable the mobility service in MIPv6, the mobile node (MN) is assigned a permanent home address in its home network (HN), and establishes a connection with the communication peer, the correspondent node (CN). A home agent (HA) serves as the anchor node in the HN that tracks the network connection point (location) of a user as the user moves. Periodically, or whenever the user changes its point of attachment to the network, the user registers with the HA through binding update (BU) messages, informing of its current location and establishing a tunnel between the HA and the MN located in a visited network. In MIPv6, the HA is the centralized part of the system since it is on the critical path of both signaling and data for mobile users. Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

87

Mobile IPv6 has well-known shortcomings, such as high handoff latency, high global signaling load and scalability issues. To cope with these limitations, especially with handoff latency, several extensions including Fast Handovers for Mobile IPv6 [13] and Hierarchical Mobile IPv6 [14] have been proposed. In other relevant works the authors enhance scalability by reducing routing and mobility signaling for centralized approaches [15] and evaluate the impact of mobility anchor point selection inthe performance [16].

2.2. Mobile IP and MPLS integration In the design of next-generation mobility management protocols, one of the main goals is to develop protocols able to provision the network resources efficiently. In this perspective, there is an increasing trend towards the use of MPLS inIP-based wireless access networks to benefit from its QoS, traffic engineering and reliability capabilities [11,17]. In order to tackle the QoS provision, Mobile MPLS and FH-Micro Mobile MPLS are interesting solutions proposed in the literature. These schemes are briefly explained next. Mobile MPLS [4] was one of the first proposals to integrate the Mobile IP and MPLS protocols. It aims to improve the scalability of the Mobile IP data-forwarding process by removing the need for IP-in-IP tunneling from home agent (HA) to foreign agent (FA) using label switched paths (LSPs). In Langar et al. [4], the authors propose FH Micro Mobile MPLS to overcome some limitations of Mobile MPLS. In this scheme the fast handoff mechanism is considered to anticipate the LSP procedure setup with an adjacent neighbor subnet that a MN is likely to visit. The main idea behind FH Micro Mobile MPLS is to set up an LSP before the MN moves into a new subnet to reduce service disruption. Although these proposals solve some limitations related to both mobility management and resource provisioning issues, all them are derived from a hierarchical mobile network architecture such as Mobile IP that implements a centralized approach where enduser data traffic is encapsulated between a centralized mobility anchor (the HA in Mobile IPv6) and the mobile node. This means that all packets associated with a mobile node are first routed to the HA, so it becomes a single point of failure and abottleneck affecting network performance by slowing down the end-to-end packet transmission speed [18].

2.3. DMM Taking into account that the HA not only manages the mobility context but also manages routing, centralized architectures also suffer from signaling overhead and non-optimal routing. The introduction of distributed mobility management architectures is attracting attention from both industry and academia [3]. Moreover, the IETF has recently created a working group called DMM that is identifying the limitations and defining the problem statements for achieving DMM with the existing IP mobility support protocols. Traditional mobility management protocols such as MIPv6 can be decomposed into several logical functions to allow a more flexible design in a distributed way. The anchoring function (AF) is responsible for the allocation of home network prefix or HoA toan MN that registers in the network; the location management (LM) function manages and keeps track of the location of an MN; the mobility routing (MR) function intercepts packets to/from an MN’s HoA and forwards the packets based on the location information; the location update (LU) function is responsible for provisioning of MN location information to the LM function. In MIPv6, the HA typically provides the AF, MR and LM functions, while the mobile node provides the LU function. A representative proposal of DMM based on Mobile IP is Lee et al. [19]. In this work, the authors attempt to improve the performance of the mobility by distributing mobility agents (called AMA) at the edge of the access network level and the MN is served by a mobility anchor located in the serving network. When an MN moves to an adjacent network, a tunnel is created between the serving AMA and the origin AMA, located in its home network and a new address is configured in the MN. This solution creates multiple tunnels between AMAs and with a high mobility rate; system performance is critically affected by frequent registration and the maintenance of multiple tunnels. Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

88

J. CARMONA-MURILLO ET AL.

3. DISTRIBUTED MOBILITY MANAGEMENT MPLS In this section, we introduce DM3 (distributed mobility management MPLS), a new DMM architecture that is based on Mobile IPv6. The aim is to achieve an efficient mobility management with QoS support, taking advantage of both new distributed mobility management approach and MPLS features. An example of architecture for DM3 is shown in Figure 1. We assume that an MPLS domain exists in the access network between the ingress LER/egress LER. Both ingress LER (ILER) and egress LER (ELER) are the border MPLS routers that define the limits of the access network. DM3 architecture relies on the distributed mobility agent called the mobility distributed anchor (MDA). This node provides mobility management functions and is an intermediate node between the ILER and the serving ELER. The serving ELER is the egress router the MN is currently attached to. Taking into account that the HA not only manages the mobility context but also manages routing, the main idea behind the DM3 architecture is to benefit from the position of the node that manages the mobility and routing functions in order to improve the limitations of Mobile IPv6 mentioned above. The proposed architecture is based on the forwarding chain concept (set of forwarding paths). The distributed mobility anchor agent (MDA) is responsible for the LSP redirection when the MN moves to an adjacent network. This way, the LSP will be composed of a set of forwarding paths that adapt to track host mobility and localize signaling in an area close to the location of the MN. 3.1. DM3 operation The basic operation of the architecture is illustrated in Figure 2, in which the MN moves to an adjacent network, changing its point of attachment to the network. A detailed description is as follows: Initially, when the MN moves to an adjacent network, it proceeds as follows. The MN enters an overlapped area of an adjacent subnet and receives an L2 signal from the possible new base station (BS) (step 1). Next, the MN notifies the previous egress LER(PELER) of the possibility of a handoff by sending a handover initiate (HI) message that contains the new base station identifier. This information is used to obtain the new egress LER (NELER) IP address, thanks to a data structure that maintains a match between this identifier and each adjacent LER IP address (step 2). It is supposed that the MPLS access network belongs to the same service provider. Once the PELER knows the subnet which the MN is going to move, it sends a message upstream to the selected MDA, notifying a possible L3 handover, and starting the setup of a new section of the LSP from MDA to NELER (step 3) with the required QoS. In this step, the PELER also informs the MN about the NELER IP

Figure 1. Overview of the DM3 approach Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

89

Figure 2. Handover from PELER to NELER in DM3

address through an IPv6 neighbor discovery message. At this moment, a new section of the LSP tunnel could be set up so data traffic could be forwarded towards the new location of the MN (step 4). When the signal strength received from the current base station falls below a certain threshold level, the MN notifies the handoff to the PELER (step 5). Now, our mechanism responsible for minimize packet loss is started by the PELER (step 6). This threshold can be determined by handover decision algorithms that rely on received signal strength (RSS) and other link-layer parameters. The next section give details about this packet recovery mechanism. Once the L2 handover is performed, the MN initiates an L3 handover through the Mobile IPv6 registration process (step 7). The new LSP section from MDA to the new egress router will be used when the MDA is aware of the movement. This happens when the PELERstarts to return data packets to the MDA, which will be forwarded to the NELER through the new LSP section together with buffered packets according to the recovery mechanism. Finally, the MN sends the Mobile IPv6 binding update message to the mobility anchor (MDA) (step 7). The MDA will reply with a binding acknowledgment message to the MN that is located in the new subnet (step 8). The novelty of this handover procedure is the recovery mechanism, as well as the selection of the correct mobility anchor after the movement. Both processes are explained next.

3.2. DM3 recovery mechanism MPLS signaling protocols provide control mechanisms for setup, teardown, maintenance and recovery of LSPs. Extensions have been developed to work on wired networks, where paths, once established, hardly change. In mobile networks, the frequency of path disruption due to a handoff is very high. Packet loss and packet disorder are two important factors which badly affect handoff performance of DM3 due to TCP retransmissions. In TCP, the algorithms Fast Retransmit and Fast Recovery are used when packet loss and packet reordering occur. Avoiding losses and disorder, the overall performance of TCP and DM3 can be improved. In order to improve the behavior of DM3, we propose the use of a recovery mechanism that minimizes packet loss and avoids packet disorder. This mechanism is based on the use of two buffers. In this section we present the DM3 recovery mechanism. Its operation is shown in Figure 3 and is detailed as follows. Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

90

J. CARMONA-MURILLO ET AL.

(a)

(b)

(c)

(d)

Figure 3. Recovery mechanism operation The MDA nodes have a data structure of two buffers (B1 and B2) that allow them to store both in-flight packets (through the path MDA-PELER) and incoming packets. When the MN informs the PELER about an L2 handoff, this edge router does not send any more packets to the MN; instead, it sends the packets back to the MDA, as Figure 3(a) illustrates. When the first packet arrives back to the MDA (packet n in Figure 3), it tags the next packet received (n + 4) and sends it through the PELER again. The incoming packets from the PELER are buffered in B1 while incoming packets from the data path are buffered in B2 (see Figure 3b). Once the MDA receives the tagged packet (n + 4) from the reverse path, it untags it, and buffers the packet in B1 (see Figure 3c). B2 buffers the rest of the packets until the BU message arrives at the MDA. When the BU is received, all packets can be forwarded through the new section of the LSP (MDA-NELER), as Figure 3(d) shows. In order to avoid packet disorder, packets in B1 are sent before packets in B2. This way, packet loss is minimized and packets are sent in the correct order towards the new location of the MN; therefore the work required of the MN to reorder the information is significantly reduced.

3.3. Mobility functions in DM3 In DM3, mobility management is fully distributed. That means that both the data plane and the control plane are distributed. The data plane is responsible for routing packets to the corresponding peer entity, whereas the control plane maintains the mobility bindings. In the data plane, the mobility routing function is distributed to multiple locations (MDAs) and each MN is assigned to the closest light-load MDA, so the routing can be optimized, avoiding inefficient paths after various movements. In the control plane, the MDAs may signal each other, and the LM function is a distributed database, also located at MDAs that maintain the mapping of HoA to CoA. The LM functions are located in the MDA due to the improvement this provides. To perform mobility routing, the MDA needs the location information and, when an MN performs a handover, Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

91

the LM database needs to be updated. In DM3, when a handover occurs,the MDA is notified so the LM database can be updated at the same time that the mobility anchor is aware of the movement. The operation of the routing and control functions in DM3 is as follows. When a mobile node attaches to a new LER and initiates an IP communication with a correspondent node, the traffic will be anchored to an MDA. The anchor selection is an important process of the distributed control plane and it is explained in the next section. When performing handover to another network and a different MDA is selected, this movement will be shared by these two MDAs. Once the handover has been made and the previous mobility anchor has been notified of the movement, that MDA can forward packets to the new MNs. Registration update to the control function is initiated by the host. With regard to the location information, when the mobile host moves to another network the location information needs to be updated and this information may need to be disseminated among MDAs. The user data can be continuously delivered to the MN in the new location by rerouting the tunnel with a new MPLS segment between the old MDA and the NELER (passing by the new MDA). Distributed mobility management architectures require enabling dynamic mobility anchor selection mechanisms. 3.4. MDA selection process One of the most significant issues of the distributed control plane is that a special mechanism is needed to identify the mobility anchor that maintains the mobility binding of the mobile host. In DM3, every time an MN attaches to the domain or a handover occurs, it is necessary to decide which MDA should be selected for it. In DM3 architecture, mobility anchor selection depends on two main factors: the distance between the NELER and the MDA and the MDA’s load. NELER should select the closest light-load MDA. For each MDA, let Ci be its capacity in terms of number of flows supported and N the set of mobile nodes. A traffic flow originating or destined for a mobile node n ∈ N through an MDA i is given as f ni . Hence we define Li, as the load of an MDA i, as follows: Li ¼

∑n∈N ð f ni Þ Ci

where fni represents the total flow of all mobile nodes in MDA i. Let hNELERMDAi be the hop count between the LER that MN accesses and the MDA i. We suppose that each LER knows the distance, in number of hops, between it and each MDA, and there are n MDAs in this domain. Thus the output of the MDA selection can be expressed as follows: MDAselection ¼ minfLi  hNELERMDAi g 4. ANALYTICAL RESULTS In order to evaluate the performance of DM3, we analyze the cost functions of traffic routing, registration updates, total packet loss during a session and buffer size metrics. These indicators have been selected because they are the main limitations of mobility management protocols and they need to be evaluated [12]. We also compare our proposal with other works with the same objective, such as Mobile IP, Mobile MPLS, FH-Micro Mobile MPLS and host-based distributed mobility management protocol. In order to simplify the analytical study, we do not consider the cost of the process that periodically updates the link (binding update) between the mobile node and the HA (or the mobility anchor), in order to update the binding cache. We analyze the mobility behavior of the MN, keeping in mind a topology where a terminal could move to every neighboring network with the same probability. In the next section, quantitative and experimental results are presented. The parameters andvalues used for both quantitative and qualitative analysis are listed in Table 1. For the numerical analysis, the parameter values are also used in Langar et al. [4] and relative distances in hops are shown in Figure 4. Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

92

J. CARMONA-MURILLO ET AL.

Table 1. Parameter values Parameter ts tr Tad Nh su sl sd Bw Bwl Lw Lwl Pt λd hx  y

Value

Description

1000 s 5–50 s 1s ts/tr 48 bytes 28 bytes 1 kbyte 100 Mbps 11 Mbps 1 ms 2 ms 10 6 s 64 kbps Figure 4

Average connection time for a session Average stay time at a visited network Time interval between agent advertisement messages Average number of level 3 handover in a session Average size of a signaling message for record update Average size of a message for LSP establishment Average size of a data packet Bandwidth of the wired link Bandwidth of the wireless link Latency of the wired link (propagation delay) Latency of the wireless link (propagation delay) Routing or label table lookup and processing delay Transmission rate for a downlink packet Number of hops between x and y

Figure 4. Relative distances in hops in the simulated network Moreover, we define t(s,hx  y) as the time that takes a message of size s to be forwarded from x to y through the wired and wireless links. t(s,hx  y) can be expressed as follows:  

s þ Lw þ hxy þ 1  Pt t s; hxy ¼ c þ hxy  Bw where c ¼ Bs þ Lwl if x = MN or 0 otherwise. wl

4.1. Routing cost The total routing cost for a session can be defined as Rd. This value is defined by the size of the messages multiplied by the number of hops needed to forward packets from the CN to the MN or vice versa. In this section we analyze the routing cost of MIPv6, a centralized mobility management solution, and our proposed DM3, a distributed approach. As we have stated previously, centralized mobility management solutions have several drawbacks such as singlepoint of failure (HA in MIPv6) and non-optimal routing. In the MIPv6 protocol, data traffic is encapsulated between the MN and its centralized mobility agent that forwards packets to the corresponding CN. To perform routing, DM3 distributes mobility routing functions in the access network to deliver the traffic in an optimized way. The values of routing cost for each solution are as follows: Rd ðMIPv6Þ ¼ ðsu þ sd Þ  hCNHA  N h þ ðst þ sd Þ  hHAMN  N h Rd ðDM3Þ ¼ ðsu þ sd Þ  hCNILER  N h þ ðsl þ sd Þ  hILERMDA  N h þ þðsl þ sdÞ  hMDAMN  N h 4.2. Signaling cost The total signaling cost of registration update for a session can be defined as Cu. This value depends on the traffic load when signaling messages are sent; i.e. the cost depends on the size of the signaling Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

93

messages and the number of hopsin every level 3 handover process during the time interval that MN communication remains active. Therefore, the cost is defined by the size of the messages multiplied by the number of needed hops. Each movement between neighboring subnets implies sending several signaling messages. In Mobile IP and Mobile MPLS cases, the registration update is needed with the HA, whereas in FH the update is local with the root of the domain, called LERG. The distributed approaches, such as DM3 and host-based DMM, update their movements with the distributed anchor, located closer to the location of the mobile node. In host-based DMM, the mobility anchor is called AMA and is located in the access router, so AMA is the first IP-capable router for the MN. Apart from signaling the mobility management protocol, some proposals also add the cost of LSP procedure setup. This is the case of Mobile MPLS, FH-Micro Mobile MPLS and DM3. This way, we obtain the following values for the signaling cost when the registration update process occurs: C u ðMobileIPÞ ¼ 2  su  hMNHA  N h Cu ðMobileMPLSÞ ¼ 2  su  hMNHA  N h þ 2  sl  hFAHA  N h C u ðFH  MMMÞ ¼ 2  su  hMNLERG  N h þ 2  su  hFAFA  N h þ þ2  sl  hFALERG  N h Cu ðDM3Þ ¼ 2  su  hMNMDA  N h þ 2  sl  hMDANELER  N h Cu ðHB  DMMÞ ¼ 2  su  hAMAi AMAiþ1  N h þ Nh

þ∑ ð2  st  hAMAi AMANh Þ i¼1

4.3. Packet loss during a session Packet loss during a session (Ploss) is defined as the sum of lost packets per MN during all handoffs. Apart from FH and DM3, in the other schemes, all in-flight packets will be lost during the handoff disruption time due to the lack of any buffering mechanism. Therefore, the value Ploss for each proposal is 

  1 T ad þ T c ðMobileIPÞ λd N h 2    1 T ad þ T c ðMobileMPLSÞ λd N h Ploss ðMobileMPLSÞ ¼ 2 Ploss ðFH  MMMÞ ¼ t ðsu ; hMNFA Þλd N h Ploss ðMobileIPÞ ¼

Ploss ðDM3Þ ¼ t ðsu ; hMNPELER Þλd N h    1 T ad þ T c ðHB  DMMÞ λd N h Ploss ðHB  DMMÞ ¼ 2 where Tc is the average time of the handover completion, which is defined as the sum of three terms: interruption time, establishment time and Tinter/2. 4.4. Buffer size As stated in a previous section, both FH and DM3 minimize packet loss during handover so they need a buffer to store in-flight packets. In DM3, the buffer is located at the MDA and its size depends on the time needed for the recovery mechanism to store the packets in the correct order. In the FH-Micro Mobile MPLS scheme the buffer is located in the LER/FA node (a border router) and the buffer size depends on the time it takes a message to be forwarded from the MN to the FA plus the time it takes for a message to be forwarded to the FA to the next FA (where the MN is finally attached). In the other Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

94

J. CARMONA-MURILLO ET AL.

mechanisms, all in-flight packets are lost during the handoff time due to the lack of any buffering mechanism. Therefore, the buffer size requirement (Bsize) is listed as follows:   1 T ad þ t ðsu ; hMNFA þ hFAFA Þλd 2   1 T ad þ t ðsu ; hMDAPELER þ hPELERMDA Þλd Bsize ðDM3Þ ¼ 2

Bsize ðFH  MMMÞ ¼

5. PERFORMANCE EVALUATION In this section, numerical results are presented in order to examine the behavior and performance of the different schemes. Moreover, we present the results of a testbed where we have evaluated the location of the distributed nodes in the access network. 5.1. Quantitative results Figure 5 depicts the routing cost of forwarding data traffic during a session. The graph shows the resulting values of both the standardized mobility solution (MIPv6) and a distributed solution such as DM3. As can be observed, DM3 allows the traffic to be anchored closer to the mobile node and provides lower routing cost values. In MIPv6, all packets are routed through a centralized node and this often results in longer paths from MN to CN. Control traffic functions in a similar way. Figure 6 shows the comparison of signaling cost of registration update vs. resident time. In this case, the Mobile MPLS scheme is the costliest proposal due to the requirements to establish a complete LSP tunnel from MN to HA apart from the specific mobility signaling. Conversely, DM3 uses the resources in the MPLS access network efficiently since it distributes the HA mobility functions in MDA nodes, not overloading links and nodes near the ILER. This way, DM3 can significantly reduce the registration cost particularly when the MN handoff frequently (i.e. the resident time in each subnet is short). The introduction of MDA nodes in the MPLS domain allows a reduction in the signaling exchange.

Figure 5. Routing cost of data traffic in MIPv6 and DM3 Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

95

Figure 6. Signaling cost of registration updates This also happens with the host-based DMM protocol and both distributed protocols reduce the signaling cost of registration update significantly. Figure 7 shows the amount of packet loss during the whole connection session for the different approaches. These results show the large difference between the proposals which have buffering

Figure 7. Packet loss during a session Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

96

J. CARMONA-MURILLO ET AL.

mechanisms and those which do not. Mobile IP, Mobile MPLS andhost-based DMM have the largest amount of packet loss due to the lack of a buffering mechanism during handover disruption time. In order to minimize the packet loss, FH-Micro Mobile MPLS and DM3 include mechanisms to reduce the amount of losses. In both solutions, the previous serving router of the MN is the one responsible for initiating the buffering mechanism. DM3 also achieves an ordered delivery thanks to the recovery mechanism described in previous sections. Figure 8 shows the buffer size vs. the bandwidth of the MPLS access network. In this graph we can observe that from 300 Mbps the size of the buffer maintains rather stable values around 1 ∼ 1.25 kb. 5.2. Experimental results In order to complement the analytical results presented in the previous section, we have built a testbed and a set of experiments to analyze the performance of the access network depending on the location of the distributed mobility anchors. This point is one of the most important decisions in the design of a mobility management solution. In this testbed, the access network is an MPLS domain and the mobility anchors are located at different distances (hops) from the access routers that currently serve the mobile node. The routers of the testbed are Cisco K9/1921 with traffic engineering (TE) capacities. Figure 9 shows the layout of the testbed. Both handover latency and packet loss are evaluated in the experiments. Note that L2 information is not given to the routers. In order to assure the reliability of the results, several experiments were performed using UDP traffic at different rates (1, 30 and 50 Mbps) and with different packet sizes (100, 600 and 1470 bytes). All the experiments shown in this section were made with a cell resident time = 100 s; that is, one handover is performed each 100 s. The example in Figure 10 shows one of the results from the testbed. In this case, three executions were made with apacket size of 600 bytes and at different throughput (1, 30 and 50 Mbps). The handover occurs at second 30 approximately and the mobility anchor is located at 1 hop from the old serving node of the mobile node. In the following sections, the experimental results of handover latency and packet loss during the movement of a user with different configurations of throughput, packet size and location of the distributed mobility anchor are given.

Figure 8. Buffer size Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

97

Figure 9. Testbed scenario

Figure 10. Testbed executions at different throughputs 5.2.1. Handover latency Figure 11 shows the impact of the distance from the mobile node to the mobility anchor in the handover latency. The results outline the benefits of locating the mobility anchor closer to the user. If the mobility functions are distributed to anode placed at 1 hop from the edge of the access network instead of 4 hops, our experiments outline that the overall handover latency is reduced, on average, by 11.95%. With this experiment, we demonstrate that the topological location of the mobility anchor affects the handover performance. In centralized approaches, the distance between the user and the HA is higher so the handover latency would be further increased. Distributed mobility management approaches can reduce this time by locating the mobility anchor closer to the MN. 5.2.2. Packet loss During the handover disruption time, the mobile node cannot receive IP packets. Accordingly, the topological location of the mobility anchor affects not only the handover performance but also the packet loss. Without any buffering mechanism, data packets sent from the CN to the MN will be lost while the MN changes its point of attachment to the network. In this experiment, we analyze the impact of the location of the mobility anchor on the packet loss. As we mentioned previously, the experiments were made at different bandwidths and using different packet sizes. Moreover, one handover occurs each 100 s. The graphs shown in Figure 12 illustrates the Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

98

J. CARMONA-MURILLO ET AL.

4

Time (sec.)

3

2

3,095

3,225

3,352

3,465

Dist=2

Dist=3

Dist=4

1

0 Dist=1

Distance (Number of hops)

Figure 11. Handover latency results of the experiments. In Figure 12a, we canobserve the packet loss where the mobility anchor is located at different hops based on the throughput, and in Figure 12b the amount of loss is based on the packet size. A complete summary of the results is shown in Table 2. The results shown in the graphs are grouped by the distance to the mobility anchor. Similar to the handover latency experiment, a distributed mobility management solution can reduce the packet loss rate. Moreover, due to the nature of the mobility, where periodically packets are lost, buffering techniques can reduce this limitation. Throughput 1 Mbps 30 Mbps 50 Mbps

10

5

Packet size 1470B 600B 100B

20

Packet loss (%)

Packet loss (%)

15

15

10

5

0

0 Dist=1

Dist=2

Dist=3

Dist=4

Dist=1

Distance (Number of hops) (a)

Dist=2

Dist=3

Dist=4

Distance (Number of hops) (b)

Figure 12. Impact of the location of the mobility anchor on packet loss (%) under different conditions: (a) at different throughputs (1, 30 and 50 Mbps); and (b) at different packet sizes (100, 500 and 1470 bytes) Table 2. Packet loss percentage during experiments Throughput 1 Mbps 30 Mbps 50 Mbps

Packet size

Dist. = 1

Dist. = 2

Dist. = 3

Dist. = 4

100 B 600 B 1470 B 100 B 600 B 1470 B 100 B 600 B 1470 B

13.1 6.3 3 19 8.7 6.6 24 9.2 7

13 3.6 2.7 17 7.4 5.4 22 6.7 6.2

12 3.4 2.5 14.3 5.4 5.3 20 5.7 5.9

8.7 2.3 2.3 12 6 5 18 5.1 5.3

Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

DM3

99

In these buffers, packets can be stored during a short period of time. Our DM3 solution proposes a buffering mechanism in order to minimize the packet loss during handovers.

6. CONCLUSION In this paper, we proposed a fully distributed mobility management architecture called DM3 (distributed mobility management MPLS) to solve the limitations of existing IP mobility management solutions relying on centralized mobility anchors. DM3 suggest the inclusion of distributed mobility anchors in the access network, close to the user that they are serving. This way, the path between the MN and the correspondent node follows an LSP path composed by a set of forwarding chains. We proposea flexible access network where data plane routing is optimized with the location of the mobility anchors in order to build an efficient routing path to track the MN’s movement. The location of the mobility anchor is a crucial decision in DMM architectures. We have built a testbed to analyze the impact of the location of the mobility anchor on the performance of handover latency and packet loss. Our experimental results demonstrate the benefits of distributing the mobility anchors close to the user. This is one of the reasons why mobile networks are moving towards flatter designs. Moreover, a performance comparison between our proposal and existing protocols was made. The results show that our mechanism solves the limitation of centralized mobility proposals and offers a small registration update cost and minimizes the packet loss rate.

REFERENCES 1. Perkins C, Johnson D, Arkko J. Mobility Support in IPv6. RFC 6275, July 2011. 2. Gundavelli S, Leung K, Devarapalli V, Chowdhury K, Patil B. Proxy Mobile IPv6. RFC 5213, August 2008. 3. Zuniga JC, Bernardos CJ, de la Oliva A, Melia T, Costa R, Reznik A. Distributed mobility management: a standards landscape. Communications Magazine, IEEE 2013; 51(3): 80–87. 4. Langar R, Bouabdallah N, Boutaba R. A comprehensive analysis of mobility management in MPLS-based wireless access networks. IEEE/ACM Transactions on Networking 2008; 16(4): 918–931. 5. Samdanis K, Aghvami AH. Qos provision in wireless access networks: a routing perspective considering mobility. International Journal of Network Management 2009; 19(5): 445–456. 6. Chan H, Liu D, Seite P, Yokota H, Korhonen J. Requirements for distributed mobility management. Internet draft, August 2013. 7. Tipmongkolsilp O, Zaghloul S, Jukan A. The evolution of cellular backhaul technologies: current issues and future trends. Communications Surveys and Tutorials, IEEE 2011; 13(1): 97–113. 8. Ghebretensae Z, Harmatos J, Gustafsson K. Mobile broadband backhaul network migration from TDM to carrier Ethernet. Communications Magazine, IEEE 2010; 48(10): 102–109. 9. Passas N, Salkintzis AK, Wong KD, Varma VK. Architectures and protocols for mobility management in all-IP mobile networks [guest editorial]. Wireless Communications, IEEE 2008; 15(2): 6–7. 10. Langar R, Bouabdallah N, Boutaba R, Sericola B. Proposal and analysis of adaptive mobility management in IP-based mobile networks. IEEE Transactions on Wireless Communications 2009; 8(7): 3608–3619. 11. Garroppo RG, Giordano S, Tavanti L. Network-based micro-mobility in wireless mesh networks: Is MPLS convenient? In Global Telecommunications Conference: GLOBECOM 2009, IEEE, 2009; 1–5. 12. Kong K-S, Lee W, Han Y-H, Shin M-K, You HR. Mobility management for all-IP mobile networks: mobile IPv6 vs. proxy mobile IPv6. Wireless Communications, IEEE 2008; 15(2): 36–45. 13. Koodli R. Mobile IPv6 Fast Handovers. RFC 5568, July 2009. 14. Soliman H, Castelluccia C, ElMalki K, Bellier L. Hierarchical Mobile IPv6 (HMIPv6) Mobility Management. RFC 5380, October 2008. 15. Samdanis K, Friderikos V, Aghvami AH. Autonomic hierarchical reconfiguration for wireless access networks. Journal of Network and Computer Applications 2009; 32(3): 630–641. 16. Pack S, Kwon T, Choi Y. A performance comparison of mobility anchor point selection schemes in hierarchical mobile IPv6 networks. Computer Networks 2007; 51(6): 1630–1642. 17. Xie K, Wong VWS, Leung VCM. Support of micro-mobility in MPLS-based wireless access networks. IEICE Transactions on Communications 2005; E88-B(7): 2735–2742. 18. Shin D-H, Moses D, Venkatachalam M, Bagchi S. Distributed mobility management for efficient video delivery over all-IP mobile networks: competing approaches. Network, IEEE 2013; 27(2): 28–33. 19. Lee J-H, Bonnin J-M, Lagrange X. Host-based distributed mobility management support protocol for IPv6 mobile networks. In IEEE 8th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), 2012; 61–68.

Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem

100

J. CARMONA-MURILLO ET AL.

AUTHORS’ BIOGRAPHIES Javier Carmona-Murillo received his computer science engineering degree in 2005. He graduated with honors from University of Extremadura, Spain, where currently he works as a junior lecturer and is a Ph. D. candidate at the Department of Computing and Telematics System Engineering. His research is focused on QoS provision in mobile networks and IPv6 mobility. José-Luis González-Sánchez He received his Engineering degree in Computer Science and his Ph.D degree in Computer Science (2001) at the Polytechnic University of Catalunya, Barcelona, Spain. He has published many articles and books and directed research projects related to computing and networking. Currently, he is the general manager of the Fundation COMPUTAEX and the Center CénitS. David Cortés-Polo He got his BS and MS degree in Computer Science at University of Extremadura, 2006. Now, he is a Ph.D. candidate at Telematics Engineering Area (UEx). His areas of interest are IPv6 Mobile Networks, MPLS-TE and QoS support. Currently, he is the Network administrator of the Fundation COMPUTAEX and the Center CénitS. Francisco-Javier Rodríguez-Pérez received his MEng in Computer Science Engineering at the UEx in 2000, where he is currently a Collaborator Professor of the Computing Systems and Telematics Engineering Department, and a Ph.D candidate of the GITACA research group. His research is mainly focussed on QoS and traffic engineering, packet classification and signalling development over IP/MPLS systems.

Copyright © 2013 John Wiley & Sons, Ltd.

Int. J. Network Mgmt 2014; 24: 85–100 DOI: 10.1002/nem