Keywords. IPv6 network; network topology auto-discovery; IPv6 address; IPv6 stateless address autoconfiguration

Transcription

1 Research of the Topology Auto-discovery Approach in the IPv6 Access Network Shen Zengwei School of Computer Science and Engineering Beijing University of Aeronautics and Astronautics Beijing, ,P.R.China Zhou Gang School of Computer Science and Engineering Beijing University of Aeronautics and Astronautics Beijing, , P.R.China Abstract. For processing effective IPv6 network s management, it is crucial to carry out network topology discovery based on the characteristics of IPv6 network. Most research at present mainly focused on the approaches of IPv6 backbone network s topology discovery, but there s hardly any exploration to the approach of IPv6 access network s topology discovery. A strategy of topology discovery in the IPv6 access network is to be proposed in this paper. The correctness and the effectiveness of our approach are also to be analyzed through some tests in the end. Keywords. IPv6 network; network topology auto-discovery; IPv6 address; IPv6 stateless address autoconfiguration I. Introduction As the most important protocol in the TCP/IP suites, IP Protocol carries the responsibility for routing and addressing in the Internet. But because of the IP address s lack, the IP version 4 (IPv4) widely used at present can t meet the demand of the Internet s future development. So IETF (Internet Engineering Task Force) began to make research on finding the IPv4 s successor in In 1994, the new IP Protocol was named as IP version 6 (IPv6); then, the kernel protocol suite of IPv6 became IETF s official criterion in 1998 [1].IPv6 is probable to take the place of IPv4 before long. China attached a lot of importance to the development of IPv6. As the IPv6 backbone of CNGI (China Next Generation Internet), Cernet2 began to work in China also attached a lot of attention to the cooperative plan for the next generation Internet with Japan, and the WAN which connects the two countries has been accomplished. Topology can be defined as a graph whose nodes represent the network elements and edges stand for the physical or the logical interconnections between the elements. Topology discovery technology is an important way to describe the network s action. Topology graph can describe the architecture and the development trend of the network in a visible way. It is also fundamental for the IPv6 network management solution in the future. Precise topology data is crucial for the network management system to do the performance management, the fault management, etc. Because of the different characteristics between IPv6 and IPv4, the traditional IPv4 topology discovery technology is not appropriate for IPv6. For example, IPv6 networks usually have huge address space, so any iterative method used for the IPv4 devices detection is infeasible. What s more, most next hop entries in the IPv6 route table just can be addressed on the same link, so the discovery server can t find the new devices via next hop address. This paper is dedicated to the approach of topology discovery in the IPv6 network to meet the IPv6 network management s requirements. This work has been funded by 973 Project of China (2005CB321901) The paper is organized as follows. Firstly, we will discuss the architecture of IPv6 network and IPv6 address (Section II). And then, we will summarize the IPv6 topology discovery approach in brief (Section III). Based on the above know-ledge, we ll explain the key points in our solution (Section IV). In the end, we concluded with the discussion about the implementation (Section V) and the test result for our solution. II.IPv6 address and the IPv6 access network A. Format of the aggregatable global unicast address If the IPv6 hosts want to communicate with each other on the Internet, they should have an aggregatable global unicast address which has 64 bits prefix defined in RFC2374. The prefix is

2 defined hierarchical, including 3bits prefix flag, 13bits TLA (Top Level Aggregate)ID, 8 bits reserved field, 24 bits(next Level Aggregate)ID and 16 bits SLA(Site Level Aggregate)ID, as shown in Fig.1.They are used to identify ISP s different levels. Figure 1. Prefix of Aggregatable Global Unicast Address IPv6 supports stateless address configuration of global unicast address [2]. When the IPv6 nodes receive the router advertisement which provides the 64 bits prefix, it will be added up to the EUI-64 interface ID, the 128 bits aggrega-table global unicast address will be obtained. B. IPv6 access network According to ITU G.902, the access network is the system composed of the cable and the equipments from ISP s access switches to the end users network interface. Because a site doesn t have any down stream network and its topology structure can be organized by the network administrator independently, it can be seen as the basic unit of IPv6 access network. A site can comprise local links at most; the IPv6 nodes which share the same physical link are on the same local link [3]. If a host has the IPv6 stack, it will have the local link address whose prefix is FE80::/64. The local link address can only be used in the communication between the nodes on the same link, because the routers won t forward the packets which have local link address as their destination address either. III. IPv6 topology discovery approaches The process of topology discovery can be concluded as collecting different nodes configuration information through making use of types of protocol and calculating the inter-connections between the nodes according to the graph theory. The aims and the approaches of topology discovery in the backbone network or in the access network are quite different. In the backbone network, it covers a wide range, uses the generic protocols such as UDP and ICMP, so the result couldn t be precise. In the access network, it covers a much narrower range, network management protocols such as SNMP is widely used and more attention is paid to the precision and the real-time ability. So the topology discovery approaches in backbone and in access network should be researched respectively. Current research of the IPv6 network topology discovery in the world mainly focuses on the backbone topology between different TLAs. IPv6 Resource Centre of UK and National Technical University of Athens have been representing their backbone topology diagrams between the TLAs in 6Bone independently [4]. And on Mar 4th 2005, CAIDA began to put out the interconnections diagrams between the ASs [5] ; it may be a new evolvement of IPv6 backbone network s topology discovery. I.Astic and O. Festor of LORIA-INRIA Lorraine proposed the hierarchical topology discovery approach in They suggested a kind of two-level architecture including a local agent (LA) on every link [6], a global manager (CM) somewhere in the network and a protocol to exchange data between them as shown in Figure 2. Figure 2. Hierarchical topology discovery in IPv6 network. Each LA manages only one link, and its mission is to discover every node s information that can be found. The LA s algorithm can be active or passive. An active algorithm sends ICMPv6 or other

3 requests in order to collect specific information. A passive algorithm listens to all the information going through the link only. CM s job is to receive all the information coming from the LAs, to correlate them all together and to build the topology. CM uses the UDP message to communicate with the LAs. IV. Key points of topology discovery in IPv6 access network The network management technology s research in the IPv6 network hasn t been very mature yet. Because IPv6 hasn t been widely used in the commercial applications, the need of the effective IPv6 network s management is not so exigent, and the MIB related to IPv6 hasn t been well supported by the IPv6 devices manufactures either. The key points of the IPv6 access network topology discovery system include the functions of local node s discovery, the double stack information collection and the path information collection. They are going to be introduced in detail. A. Local nodes discovery In the IPv4 network, in order to determinate all the devices alive in the given subnet, the topology discovery server usually sends exhaustive ICMP echo requests to all the possible IPv4 addresses. But IPv6 s huge address space prevents using this method, because IPv6 address has 64 bits interface ID, it is unacceptable to use an iterative way to list all the 2^64 possible address. Fortunately, several kinds of reserved multicast address were defined in RFC2373, including FF02::1 (all the IPv6 nodes on this link) and FF02::2 (all the IPv6 routers on this link). Every node (including hosts, routers) on the link will accept the packets whose destination are FF02::1, but only the routers will process the packets whose destination are FF02::2. So if sending ICMPv6 echo request to FF02::1, all the IPv6 nodes alive can be determinate; if sending ICMPv6 echo request to FF02::2, all the IPv6 routers on the local link will answer. The source address of the packets which the local nodes discovery module memorized is every node s local link address whose last 64 bits is the interface ID conforming to the EUI-64 format. Sending ICMPv6 router solicitation to FF02::2, the router advertisement from the local router will designate the subnet prefix. Add it to the interface ID, one of the nodes stateless auto-configuration addresses can be found. The local nodes discovery module follows this algorithm: 1. Send an ICMPv6 echo request to the all-nodes multicast address on local link FF02::1. 2. Arm a timer 3. Wait for an interruption. 4. If a message arrived, 4.1. If the packet is an ICMPv6 echo reply, memorize the source address received, 4.2. Get the MAC address from the node s local link address. 5. If the timer ended, 5.1. Send an ICMPv6 echo request to the all-routers multicast address on this link FF02:: Arm a timer, 5.3. Wait for an interruption, 5.4. If a message arrived, If the packet is an ICMPv6 echo reply, memorize the source address received, Mark its type to be router, 6. If the timer ended, 6.1. Send an ICMPv6 router solicitation to the all-routers multicast address FF02:: Wait for an interruption If the packet is a router advertisement, save the prefix of the subnet. 7. If the prefix is known, try to compose the IPv6 auto-configured addresses. 8. Find node s string names associated with local link addresses. B. Double stack information collection In the interim of IPv4 to IPv6, it is common for the devices to have both of the IPv4 and the IPv6 stacks. The topology discovery service should have the capability of recognizing them on such devices. In order to find out the corresponding IPv4 address to the IPv6 address, our system has to scan the IPv4 address space on this link. Because there is relationship between IPv6 local link address and the MAC address, nodes MAC address on the link can be got from the local discovery module. After listing all the address in the IPv4 subnet space, there will be some mapping information between the IPv4 address and the MAC address in the server s ARP buffer. Because every interface has a unique MAC address, the IPv4 and the IPv6 address which map the same MAC address are of the same double stack device. The algorithm can be described as: 1. Get the MAC address from local nodes discovery nodes. 2. Learn LA s local IPv4 address and subnet mask.

4 3. Calculate the IPv4 address space on this link. 4. Send IPv4 ICMP echo reply to all of the possible IPv4 address on this link. 5. Get the IP-MAC information from local ARP buffer. 6. Find the IPv4 and IPv6 address which can match the same MAC address. C. Path information collection The information about the connection between the local network and the backbone should be got via the topology discovery service too. If call the traceroute6 function which sends the UDP packets with the increasing hop limit value, according to the Time Exceeds Message sent by the routers, the path from the local link to the CM can be got, then the local link s access point can be found. It can be described as: 1. Call the traceroute function, the CM is the destination. 2. Arm a timer. 3. Wait for an interruption If the packet is an ICMPv6 time exceeds packets, memorize the source address. 4. If timer ended. 5. Send all of the information to CM. V.Implementation experience National Lab. of Software Development Environment has implemented the IPv6 hierarchical topology discovery system. Local link s topology discovery subsystem is independent from the backbone network s subsystem; they comprise of the LA (Local Agent) and CM (Central Manager) architecture as shown in Figure 3. Figure 3. Architecure of the IPv6 topology discovery system The backbone s topology discovery subsystem is used to detect the interconnections between the backbone routers, and it is deployed on the CM; LA is in charge with the collection of local links topology information; CM create the global topology graph on the basis of the local links and the backbone network s topology information. Our system s architecture looks the same as LORIA-INRIA Lorraine s discovery system introduced before. But there s some differences between them, such as the imple-mentation and the network range they covered. First of all, LORIA-INRIA Lorraine s system is used to manage their LAN; but our system s scanning range includes the WAN. Secondly, the backbone information in LORIA-INRIA Lorraine s system is got from the LAs; but in our system, the traceroute destination is analyzed from the Internet s BGP route table, so it will cover the networks in which there s no LA, the gateways which the LAs found will be attached to the current topology graph. What s more, LORIA-INRIA Lorraine s LA uses some passive algorithm to listen in the link to avoid additional traffic; but in our system, in order to insure the topology graph s real-time ability, we collect all of the information in the local link by sending certain requests completely. If some information can t be got from the response directly, we will calculate it through some rules. A. CM s and LA s structure In this paper, we mainly discuss the LA s design and implementation. But in order to explain the overall structure of our system, we will introduce the CM in brief. CM is in charge with receiving the message from the LAs, and then these local links topology will be correlated with the graph which has been got. In order to collect the backbone topology, CM will look into its gateway s BGP route table to find out the possible IPv6 routers in the WAN. And

5 these router items will be the destination of the traceroute function; its result will be a series of lines, and then they will be interlaced with each other to compose a network by seeking their cross points. The proposal about the CM can be got from The LA comprises three modules: local discovery, double stack information collection and path information collection and communication as shown in Fig. 4. Figure 4. Modules of the IPv6 topology discovery system The local discovery module discovers the IPv6 nodes via sending the ICMPv6 echo request and router solicitation packets to the multicast address FF02::1 and FF02::2. The nodes stateless autoconfiguration address can be got from the information of local link address and router prefix. And then, this module will get the devices string name by DNS requests. The tuples of <IPv6 address, MAC address> got from the local discovery module will be passed to the double stack information collection module. After listing all the address in the IPv4 subnet space, we can get the pair of <IPv4 address, MAC address>. Through finding out the pairs with the same MAC address, the pairs of <IPv4 address, IPv6 address > will be got at last. These information will be passed to the path information collection and communication module. After calling the traceroute function to find out the path between LA and CM, the local nodes and the path s information will be sent to CM encapsulated in the SNMP trap packets [10] and TCP packets. B. Network environment The practical topology discovery system is shown in Figure 5. LAs are deployed in the subnets 2001:251:1E01: 14:2::/80, 2001:251:1E01:14:3::/80, 2002: D219:8538:4::/64, and 3FFE:3240:8006::0/64. Subnets 2001:251:1E01:14:2::/80 and 2001:251:1E01: 14:3::/64 are connected to the China and Japan Cooperative IPv6 Testing Network via static routes; subnet 2002:D219: 8538:4::/64 is connected to the global IPv6 Internet through 6to4 tunnel broker; 3FFE:3240:8006::0/64 is connected to Cernet s backbone via BGP. They have different route distance to the CM, and these four subnets access the backbone network through different methods such as tunnels, static routes or dynamic routes respectively, so we choose them as the experiment environment for the LAs. CM is on a double stack host which connects to the Cernet s AS border routers directly, and it can communicate with all of the LAs.

6 Figure 5. The experiment environment of the topology discovery system C. System measurement These four subnets prefix and the of nodes in them are shown in Table 1. Because IPv4 subnet s space on the links affects the topology discovery time, it will be listed below too. Table 1. The information of the experimental IPv6 network subnet IPv4 address space of nodes A 2002:D219:8538:4::/ B 3FFE:3240:8006::0/ C 2001:251:1E01:14:3::/ D 2001:251:1E01:14:2::/ LA is running on the host with Redhat Linux9 whose memory size is 128 megabytes and CPU is PII 333 MHz. The nodes on the same link are interconnected with 100Mb/s Ethernet. Every LA is on one host in these networks, so it shouldn t bring any negative effects to the nodes and the network. We did some measurement to validate LA s consumption. The percentage of the CPU time and the memory size used can explain the influence to the host s performance, and the input and output flux generated by LA will explain its influence to the network. The measurement can be divided into three steps: 1. Start up the local agent; 2. Use the shell command of ps -eo pid,tt,user,fname, %mem,%cpu to get the percentage of CPU time and the memory size that the LA process used; 3. Get total bytes of the ICMP, ICMPv6 and UDP packets the agent sent in topology discovery with tcpdump, and then we can get the average flux from the result. The result is shown in Table 2: subnet Table 2.The resource consumed by the local agent percentage of memory size percentage of CPU time discovery time (seconds) input flux (KB/S) output flux (KB/S) A 0.2% 0.2% B 0.1% 0.1% C 0.1% 0.1% D 0.1% 0.1% The process of LA hardly consumed any CPU and memory resource. On the other hand, the average input and output flux were below 1KB/s, which almost can be omitted compared with the bandwidth of 100Mb/s. The result proves that the LA has little influence on the host and the network. In order to check whether LA can reflect the topology change in time, it is necessary to do the discovery delay s test. In the test, we added and reduced the IPv6 devices on the link manually and calculated the time until the topology graph changed. Let the time when the topology graph was created to be t2; the time when the practical change happened in the network to be t1; the of nodes increase and decrease to be n. The average delay d is going to be: t2-t1 d= n (1)

7 The result is shown in Table 3. Table 3. Delay of the topology data updating of the local agent subnet nodes increase nodes decrease increase increase measure d average delay (seconds ) decrease Decrease measure d average delay (seconds ) A B C D Because there isn t any commercial IPv6 network management software available, we have to compare the result with HP Openview s performance guide-line, as shown in Table 4 [7]. Table 4. Delay of the topology data updating of Openview delay of devices off-line discovery(seconds) discovery delay of devices startup (seconds) OpenView We can see LA s delay is much shorter than Openview, and it can almost reflect the access network s topology change in time. VI. Conclusion With the IPv6 network s fast development, it becomes a latent requirement to manage the IPv6 network efficiently. This paper proposed a solution of how to do topology discovery in the IPv6 access network according to the network s structure. And then, we accomplished the system s design and the software s implementation. In the end, we did some measurement to prove our system can meet the demand of the access network s topology discovery in a certain extent. Reference [1] Silvia Hargen. IPv6 Essentials. O'Reilly Media, Inc.2002 [2] S.Thomson.IPv6 Stateless Address Autoconfiguration(Track of RFC1971). IETF RFC2462, 1998 [3] Sam Brown. Configuring IPv6 for Cisco IOS. Beijing: Publishing House of Electronics Industry, ~36, 89~93 [4] [5] [6] O.Festor. A Hierarchical Topology Discovery Service for IPv6 Networks, France: LORIA-INRIA Lorraine 2002 [7]Du Yushui, Zhou Gang, etc. Strategy for updating the data of automatic topology discovery.beijing: Journal of Beijing University of Aeronautics and Astronautics