A DHCPv6 Based IPv6 Autoconfiguration Mechanism for Subordinate MANET

2008 IEEE Asia-Pacific Services Computing Conference A DHCPv6 Based IPv6 Autoconfiguration Mechanism for Subordinate MANET Shubhranshu Singh Advanced Technology Division Samsung India Software Operations Shubranshu@gmail.com Ashutosh Bhatia Advanced Technology Division Samsung India Software Operations ashutosh.78@samsung.com Abstract In order to communicate among themselves and with the devices on the Internet, a Mobile Ad-hoc NETwork (MANET) node needs to configure its interface(s) with IP address(es). MANET is a multi-hop network often consisting of mobile devices such as mobile phones, PDAs, laptops with wireless interface(s). Due to some of the MANET inherent characteristics such as mobility, multi-hop and adhoc, manual configuration of IP address(es)/prefix(es) is not desirable and practical. The Dynamic Host Configuration Protocol version 6 (DHCPv6) is Internet Engineering Task Force defined standard mechanism to autoconfigure IPv6 address in a stateful manner. This paper discusses subordinate MANET scenario and applicability of DHCPv6 in such scenario. The paper then proposes a novel approach to dynamically configure DHCPv6 relay agents in order to make DHCPv6 message exchange possible between MANET nodes (DHCPv6 clients) and DHCPv6 server ensuring minimal control overhead. Proposed solution facilitates seamless integration of MANET with the Internet. Operation of the relay agent is kept transparent to the client and server as required by the DHCPv6 specification. The optimization helps to reduce configuration latency and signal overhead. The proposed mechanism is evaluated and analyzed using ns-2.31 over IEEE 802.11 MAC/PHY layer and Ad-hoc On-demand Distance Vector routing protocol. 1. Introduction Mobile Ad-hoc NETwork (MANET) is a widely researched and implemented technology by academia and industry. Recently the industry witnessed some deployment of the technology in different flavors which in turn further accelerated work in this field. However, there are still some challenges that need to be addressed. Towards that end recently the Internet Engineering Task Force (IETF) formed a working group to solve and standardize the problem of IPv6 address/prefix autocofiguration in such network. In MANET each node can reach not only nodes in its direct transmission range, but also distant nodes, accessible through one or more intermediate nodes. Because of mobile, ad hoc and wireless interface characteristics, the paths to a particular destination changes over time with much higher frequency than in infrastructure networks. Furthermore, the network may partition and merge again later. These MANET specific characteristics imposes new challenges on address autoconfiguration mechanism and need to be considered by any address autoconfiguration mechanism in MANET. Subordinate MANET when connected to the Internet via one or more Border Routers imposes additional requirements on the address autoconfiguration mechanism such as assigned addresses should be topologically correct and globally routable. In this paper, we consider the network topology in which all the MANET nodes can connect to the Internet through Border Router(s). Border Router (BR) is the edge router connecting MANET and the Internet. Two approaches namely stateless and stateful are widely used for address autoconfiguration. The stateless address autoconfiguration [12] mechanism allows a node to generate a tentative address randomly and then use Duplicate Address Detection (DAD) procedure to detect any duplicate address. In MANET, DAD often introduces complexities since it uses timeouts. However in MANET message delay cannot be bounded. Thus the use of timeouts cannot reliably detect the absence of a message. Such unreliability can lead to a situation where the duplicate address remains undetected as discussed in [17]. Moreover, performing MANET wide DAD would also introduce broadcast-storm due to MANET wide flooding. The stateful approach is based on client-server mechanism with centralized or distributed maintenance of configuration parameters by the server. The IETF defined Dynamic Host Configuration Protocol (DHCP) [5] is an example of stateful approach. The main purpose of our paper is to propose a novel approach for IPv6 address/prefix autoconfiguration of MANET nodes using DHCPv6, ensuring its backward com- 978-0-7695-3473-2/08 $25.00 2008 IEEE DOI 10.1109/APSCC.2008.253 237

patibility as well as minimal control overhead. The mechanism allow dynamic configuration of DHCPv6 relay agent in order to make DHCPv6 message exchange possible between MANET nodes (DHCPv6 clients) and DHCPv6 server. The approach has been designed to make sure that it is simple to implement and deploy. Another important goal of the proposal is to maintain conformance with the classical network architecture thus ensuring seamless integration of MANET with the Internet. The rest of the paper is organized as follows: Section 2 discusses related work on stateful address autoconfiguration in MANET. Section 3 discusses the IETF DHCPv6 protocol operation details. Section 4 provides architecture details of subordinate MANET deployment scenario as considered in this paper. Section 5 explores the applicability of DHCPv6 in MANET. Section 6 proposes a simple and novel approach for subordinate MANET autoconfiguration and is based on DHCPv6. Section 7 presents detail simulation results and analysis. The paper concludes with section 8. 2 Related Work Several work has been done on MANET address autoconfiguration. In [2] a detail survey of various existing address auto-configuration mechanisms in MANET has been discussed. Below related work discussion is focused only on the stateful address autoconfiguration mechanisms. The stateful autoconfiguration protocol [8] presents an address assignment scheme similar to DHCPv6. However, in this protocol instead of DHCPv6 server the border router allocates the addresses to other ad-hoc nodes and manages the allocated addresses. The main advantages of this proposal are low configuration latency and scalability due to local communication, by introducing proxy nodes. The Interface Id field of IPv6 address [6] has been sub divided into ad-hoc prefix and Host Id. The nodes configured with the same proxy node would get same prefix and create a subnet which may be multi hop in nature. However, this creates the problem related to multi-link subnet as discussed in [15]. As per the paper, proxy behavior of node is only useful when large number of nodes are working as proxy agent. However, large number of proxy nodes will create lot of signal overhead due to periodic proxy advertisement messages sent by every proxy node. Finally the border router discovery is required by the underlying routing protocol. Thus requiring new messages to be defined. [1] proposes a conflict free allocation solution by providing prefixes instead of addresses. As indicated in this paper the benefits of this protocol are: use of existing IPv6 Neighbor Discovery [9] Router Advertisement (RA) message and DHCPv6 prefix delegation mechanism [16] to configure addresses. However, The nodes configured with same RA message will have same prefix in their addresses thus creating multi-link subnet. Therefore this mechanism also creates the problem as discussed in [15]. The propagation of DHCPv6 prefix message towards the initiator node may take longer path instead of shortest one because the path information is based on previously received RA message. Normally the periodicity of RA message in IPv6 is of the order of second which will drastically increase the configuration latency of this mechanism. On the other hand reducing the periodicity of RA messages would introduce lot of signal overhead since every node is transmitting RA message. The IETF draft [14] uses IP-over-IP mechanism to provide an abstraction to IP layer. The border router considers other border routers on-link through Virtual Ethernet interface configured over their MANET interfaces. As pointed in the draft, benefit of this approach is use of existing configuration protocols such as DHCPv6 and NDP without modification. However, this proposed mechanism would introduce more complexity and overhead due to encapsulation of IP-over-IP as discussed in [13]. The major issues with this approach are fragmentation and extra overhead due to encapsulation. 3 Dynamic Host Configuration Protocol (DHCPv6) The DHCPv6 [5] is based on client-server architecture. An IPv6 node (DHCPv6 client) contacts a DHCPv6 server for allocation of IPv6 address. On receipt of DHCPv6 address, request the DHCPv6 server dynamically assigns an address from a pool of addresses based on its allocation policy. If the DHCPv6 server is not on the same link as the DHCPv6 client, it is possible to use one or more DHCPv6 relay agents. The relay agent transparently forwards the messages to a different subnet such that the request message finally reaches to the DHCP server. In short, following are the sequence of operation for address autoconfiguration using DHCPv6. Detail operation of DHCPv6 is in [5]. 1. A client sends a Solicit message (SOLICIT) to locate servers. 2. In response to a Solicit message received from a client, server sends an Advertise message (ADVERTISE) to indicate that it is available for DHCPv6 service. 3. A client sends a Request message to get IPv6 address from a specific server after choosing the server. 4. A server sends a Reply message containing assigned addresses to the client. Besides the above message exchange procedure the DHCPv6 client also sends messages for renewal of address lifetime and other configuration parameter request. In this 238

paper, we only focus on message exchange mechanism related to IPv6 address/prefix autoconfiguration. The lifetime extension and other parameter configuration can be done in a similar fashion. 4 DEPLOYMENT SCENARIO The MANET is broadly classified into two categories: subordinate MANET and Autonomous MANET. Subordinate MANET is a MANET which is connected to one or more external network(s) and where such external network(s) are imposing an addressing hierarchy scheme on the MANET. Autonomous MANET is a MANET to which no external network imposes an addressing hierarchy. An example deployment scenario of subordinate MANET, as considered in this paper, is shown in Figure 1. In this scenario, MANET is connected to the Internet via Border Router (BR). There could be single or multiple mobile or fixed border routers. Border routers have at least two interfaces. The interface that connects to the Internet is called egress interface while the interface connecting to the MANET is called ingress interface. The ingress interface is often wireless in such deployment scenario. For simplicity, scenario with single DHCP server is considered however the proposed solution would work in the scenario where multiple DHCP server exists. Moreover, the DHCP server and the border router may be co-located. This kind of deployment is very common when MANET is integrated with exiting infrastructure network. In such deployment scenario, in order to be able to communicate with the nodes on the Internet, MANET nodes need to get configuration parameters from the DHCP server. Ingress Interface BR Egress Interface Figure 1: MANET Deployment Scenario 5 APPLICABILITY OF DHCPv6 FOR MANET AUTOCONFIGURATION In the considered deployment scenario as shown in Figure 1 the DHCP server is located outside MANET and is not on the same link as MANET nodes.thus there has to be relay agents configured such that DHCP link-local requestreply messages can finally reach to the DHCP server and clients respectively. In the classical networks, relay agents are manually configured for such purposes. However, due to the inherent mobile, wireless and ad-hoc nature of MANET, manual configuration of relay agent is not desirable and practical. Therefore, these relay agents need to be dynamically configured. The border router is often a fixed node and may be statically configured as relay agent with either the server address or another relay agent that would forward the messages towards the server. To allow DHCP client messages to reach up to border router, when it is not on-link, at least one relay agent is required on client link to relay messages between client and border router. In MANET, nodes are mobile and have no predefined infrastructure or movement model. This implies that all nodes should work as DHCPv6 relay unless some optimization mechanism is used. In DHCPv6, if the relay agent does not know the unicast address of DHCPv6 server or another relay agent then it uses site scoped multicast address, ALL-DHCP-SERVER as defined as defined in [5]. Since site scoped multicasting works along the line of Multicast Listener Discovery [4] protocol, similar forwarding mechanism is required in MANET to support network wide multicast. Besides, since ALL-DHCP-SERVER multicast address is site scoped address, the message would not reach to the DHCPv6 server which is located outside MANET. Therefore, to forward client messages towards border router with minimal control message overhead, the relay agents in MANET need to be configured with border router unicast address. 6 PROPOSED MECHANISM As discussed in section 5, dynamic configuration and optimal behavior of MANET nodes as DHCPv6 relay agent are important requirements for DHCPv6 based autoconfiguration of MANET nodes. The main goals of the proposed mechanisms are thus, to dynamically configure MANET nodes as relay agent and also to allow optimal behavior as a relay agent such that the overall control message overhead is minimal. We do not modify any of the standard clientserver behavior of the DHCPv6 [5] and thus allowing the proposed mechanism to be backward compatible. The optimization procedure as proposed in this paper considerably reduces the overall control message overhead thus solving one of the challenging problems in MANET. In the proposed mechanism a client after getting its own configuration parameters from the server may function as relay agent i.e Finite State Machine (FSM) for both client and relay agent co-exist. As soon as a client gets IP address from the server, it starts functioning as a potential relay agent and decision to forward any received DHCP request 239

message is based on the probabilistic algorithm explained below in subsection 6.2. DHCPv6 relay agent relays client messages to ALL- DHCP-SERVER multicast address which is a site scoped multicast address. This requires a multicast support from underlying IP layer and results in well-known problem of multicast message overhead in MANET. In order to reduce the signaling overhead, unicast address of Border Router is used to reach the server located outside MANET. Subsection 6.1 details the protocol operation and how relay agents learn border router unicast address. 6.1 PROTOCOL OPERATION 8. The client configures its address based on received RE- PLY message and starts working as a potential relay agent. To relay any request message, it uses the border router unicast address learned from the received RE- PLY message Fig. 2 shows the sequence of message exchanges when a client performs autoconfiguration. Since we have considered only one DHCPv6 server, therefore the initial message exchange for solicitation of server is not required. (a) (b) (c) Border Router Relay Agent DHCP Client REQUEST, MULTICAST REQUEST, UNICAST REQUEST, UNICAST REPLY, UNICAST REPLY,UNICAST(Border Router Address) REPLY, MULTICAST Figure 2: Protocol Message Exchange 1. The client sends a REQUEST message to ALL-DHCP- SERVER-RELAY-AGENTS. 2. Either border router or relay agents (already configured nodes) receive this REQUEST message. 3. The relay agents relay the message to border router by unicasting it. 4. The message received at border router either directly from client or through relay agent is forwarded to DHCPv6 server. 5. The DHCPv6 server allocates the address based on its policy and sends REPLY to border router in response to received REQUEST message. 6. The border router relays the REPLY message back to relay agent or directly to client after placing its global unicast address in the REPLY message option field. 7. The relay agent sends REPLY message received from border router to client. (d) Not configured Request Multicast (e) 00 11 00 11 00 11Performing autoconfiguration 00 11 Configured (Relay Agent) Response Multicast Request Unicast Response Unicast Figure 3: Protocol Operation Fig 3 demonstrates the operation of our proposed autoconfiguration mechanism. fig 3(a) is the initial condition when no node except border router is configured. The fig 3(b) and 3(c) demonstrates the scenario when Border Router is directly reachable from node performing autoconfiguration. The fig 3(d,e and f) demonstrates the condition when some of the nodes have already been configured and a new node wants to join. The server may receive duplicate address request messages if multiple relay agents forward the same request message. Since the transaction id field of all duplicate messages is same, the server discards subsequent duplicate messages and sends REPLY message only once. As discussed before, multiple relay agents forward the same REQUEST message. This increases overall signal overhead. The situation gets worse in case of dense network when number of relay agents are comparatively larger. Below sub-section 6.2 proposes an optimization technique to deal with such situation. The mechanism is passive in nature hence does not incur any extra signal overhead. 6.2 Probabilistic Behavior of Relay Agent The relay agents work in probabilistic manner. When an already configured node receives REQUEST message from (f) 240

client it relays the message towards the border router with certain probability say Rp. Rp is the approximation of possibility of a node to work as relay agent for a particular RE- QUEST message. Higher the value of Rp higher the probability to relay the REQUEST message. It may differ relaying message with probability 1 Rp. This probabilistic relaying decreases signal overhead by dynamically reducing the number of nodes acting as relay agents for a particular REQUEST message. Rp is calculated dynamically by each node every time it receives REPLY message or relays REQUEST message. It is the ratio of number of REPLY messages received from border router to the number of REQUEST messages relayed by the particular node. If the ratio is high this means that the relay agent received most of the responses against its relayed REQUEST messages. Therefore, it should relay the REQUEST message with higher probability. The probabilistic behavior automatically takes care of node mobility as well as network density. Moreover, from implementation perspective, it does not introduce additional complexity. As the node moves from sparse network region to dense network region there will be more nodes available to function as relay agent for new joining nodes. The node will not receive REPLY messages from Border Router against most of its relayed REQUEST messages. Therefore the value of Rp will decrease and hence the probability of node to function as the relay agent. Similarly, for the node movement from dense to sparse region the probability Rp will increase. 7 Simulation Experiment To perform simulation experiment and analyze the performance of the proposed solution, we used ns-2 [10]. The primary focus of the simulation experiment is aimed to gather statistics regarding average address/prefix configuration latency and protocol message overhead. 7.1 Simulation Setup We simulated the MANET over IEEE 802.11 MAC/PHY [7] module. Simulation were performed with random way point mobility model [3]. While moving from a starting point to a randomly chosen destination the speed is kept constant at 5 m/s. After reaching destination the pause time is 5 s. Then another destination is chosen randomly and same sequence is repeated until simulation ends. Network of 25, 50, 75 and 100 nodes were simulated with area size varying from 100m x 100m to 1 km x 1km. Since the initial distribution of nodes over a fixed area also effects the configuration latency we performed simulation over multiple distribution patterns and took average performance. The transmission range is kept at 100 m. Inter arrival of new nodes is uniformly distributed in the range 0-50 s. The underlying routing protocol is Ad hoc On-Demand Distance Vector (AODV) [11], although our protocol makes no assumption about underlying routing protocol. Simulation is run until all the nodes for particular simulation scenario gets configured. 7.2 Address Allocation Latency Figure 4 shows scatter plot of address allocation latency for our simulated network. More than 80% of nodes got configured within less than 1 s. They got their addresses allocated either in first or second attempt. For few nodes configuration took considerably long time. This is because either there was no relay agent available on-link with client or REQUEST/REPLY dropped due to congestion. Figure 5 shows the effect of network size and number of nodes on average allocation latency. According to simulation results, the allocation latency increases very slowly with increase in number of nodes and network size. This shows the scalability of proposed solution with respect to number of nodes and size of network. Sometimes the average latency is less even with the increase of network size, this is because distribution of nodes over the simulated area also has major impact on allocation latency. Address Allocation Latency (seconds) 10 8 6 4 2 0 0 10 20 30 40 50 Node Number "latency" Figure 4: Address Allocation Latency Distribution 7.3 Signal Overhead Figure 6 shows the average number of control messages per address/prefix allocation. We have used only link scope multicast and unicast messages in our solution. Both kind of messages incur least amount of communication overhead and utilization of network resources compared to when site wide multicast addresses are used. The proposed algorithm allows node to learn unicast address of border router so that the relay agent can unicast the message to BR thus avoiding MANET wide flooding. Additionally, our optimization to avoid duplicate REQUEST messages due to multiple relay agents further reduced signal overhead, especially in 241

dense network. As depicted in fig 6, the number of average messages per address allocation reduces with increase in network size. As the size of network increases number of nodes per unit area (density) decreases, therefore the number of relay agents available to relay message also decreases. This in turn reduces the number of messages required to configure the node. Address Allocation Latency (seconds) Average Number of messages per address allocation 10 8 6 4 2 20 15 10 0.1x.1.2x.2.3x.3.4x.4.5x.5.6x.6.7x.7.8x.8.9x.9 1.0x1.0 5 Area (km x km) 25 nodes 50 nodes 75 nodes 100 nodes Figure 5: Average Allocation Latency 0.1x.1.2x.2.3x.3.4x.4.5x.5.6x.6.7x.7.8x.8.9x.9 1.0x1.0 Area (km x km) Figure 6: Average Signal overhead 8 Conclusion and Future work 25 nodes 50 nodes 75 nodes 100 nodes In this paper we proposed a novel mechanism for IPv6 address/prefix autoconfiguration of subordinate MANET. The mechanism allows a MANET node to dynamically configure as DHCPv6 relay agent and also ensures optimal behavior such that overall control message overhead is minimal. The solution makes use of existing DHCPv6 protocol and ensures backward compatibility. This facilitates smooth deployment and integration with existing infrastructure network. The proposed mechanism is evaluated and analyzed using ns-2.31. The result shows that the MANET nodes are able to dynamically function as relay agents and configure address(es)/prefix(es) within reasonable latency and signal overhead. In this paper, we focused only on IPv6 address/prefix autoconfiguration. The lifetime extension and other network parameters configuration can be done in a similar fashion. Besides, the considered deployment scenario could be extended by considering subordinate MANET with multiple DHCPv6 server and border routers. References [1] C. J. Bernardos and M. Calderon. A dhcpbased ip address autoconfiguration for manets. http://hdl.handle.net/106/2804,ietf. [2] C. J. Bernardos and M. Calderon. draft-bernardos-manetautoconf-survey-02, survey of ip address autoconfiguration mechanism for manets. IETF, April 2008. [3] J.B.David,A.M.David,B.J.Y.-C.Hu,andJ.Jetcheva.A performance comparison of muilti-hop wireless ad hoc routing protocols. Proceedings of the Fourth Annual ACM/IEEE International Conference on Mobile Computing and Networking, October 1998. [4] S. Deering, W. Fenner, and B. Haberman. Multicast listener discovery (mld) for ipv6. RFC 2710, Internet Engineering Task Force, Network Working Group, October 1999. [5] S. Droms, J. Bound, B. Volz, T. Lemon, C. Perkins, and M. Carney. Dynamic host configuration protocol for ipv6 (dhcpv6). RFC 3315, July 2003. [6] R. Hinden and S. Deering. Ip version 6 addressing architecture. RFC 2373, IETF, July 1998. [7] IEEE. Wireless lan medium access control (mac) and physical layer (phy) specifications. 2007. [8] D. Lee, J. Yoo, H. Kang, K. Kim, and K. Kang. Distributed ipv6 addressing technique for mobile ad-hoc networks. In Proceedings of the ACM symposium on Applied computing, 2006. [9] T. Narten, E. Nordmark, W. Simpson, and H. Soliman. Neighbor discovery for ip version 6 (ipv6. RFC 4861,IETF. [10] NS2. The network simulator ns2. http://www.isi.edu/nsnam/ns. [11] C. E. Perkins, E. M. Belding-Royer, and S. Das. Ad-hoc ondemand distance vector (aodv) routing. RFC 3561, Internet Engineering Task Force, July 2003. [12] S.Thomson, T. Narten, and T. Jinmei. Ipv6 stateless address autoconfiguration. RFC 4862, September 2007. [13] F. Templin. Subnetwork encapsulation and adaption layer. draft-templin-seal, IETF, Feb 2009. [14] F. Templin, S. Russert, and S. Yi. Manet autoconfiguration. draft-templin-autoconf-dhcp, IETF, August 2008. [15] D. Thaler. Multi-link subnet issues. RFC 4903, IETF, June 2007. [16] O. Troan and R. Droms. Ipv6 prefix options for dynamic host configuration protocol (dhcp) version 6. RFC 3633, Internet Engineering Task Force, Network Working Group, September 2003. [17] N. Vaidya. Weak duplicate address detection in mobile ad hoc networks. In Proceedings of 3rd ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 02), pages 206 216, 2002. 242