System Performance Analysis of IPv6 Transition Mechanisms over MPLS Parisa Grayeli, Dr. Shahram Sarkani, Dr. Thomas Mazzuchi 1776 G Street NW Washington DC 20052, USA The George Washington University Phone: 1-703-547-8354 Fax: 1-703-464-8882 parisag@gwmail.gwu.edu, sarkani@gwu.edu, mazzu@gwu.edu http://www.gwu.edu/ Complex Systems Design & Management (CSDM) 2011 Paris, France Abstract Exhaustion of current version of Internet Protocol (IPv4) addresses initiated development of nextgeneration Internet Protocol (IPv6). IPv6 is acknowledged to provide more address space, better address design, and greater security. IPv6 offers more IP addresses than IPv4 offers, however the two protocols are incompatible. For the two protocols to coexist, various IPv6 transition mechanisms have been developed. This research will analyze a series of transition mechanisms over the Multiprotocol Label Switching (MPLS) backbone using a simulation tool (OPNET) and will evaluate and compare their performances. The analysis will include comparing the end-to-end system performance using tunneling mechanisms such as Manual Tunnel, Generic Routing Encapsulation (GRE) Tunnel, Automatic IPv4-Compatible Tunnel, and 6to4 Tunnel between Customer Edge (CE)-to-CE routers and between Provider Edge (PE)-to-PE routers. The results are then compared against 6PE, Native IPv6, and Dual Stack all using the MPLS backbone. The traffic generated for this comparison are database access, email, File Transfer, File Print, Telnet, Video Conferencing over IP, Voice over IP, Web Browsing, and Remote Login. The performance metric includes end-to-end system delay, jitter, and throughput. A statistical analysis will be performed to compare the performance metric of these mechanisms to evaluate any statistically-significant differences between these mechanisms and to identify the superior mechanism(s). This study will be critical for those who want to implement IPv6 and are concerned about the performance of the transition mechanisms. Given the acquisition cost, schedule, risk, and technical challenges in supporting IPv6, this analysis will provide increased confidence to make informed decision and choose the correct IPv6 transition mechanism depending on the performance requirements. Keywords: IPv4, IPv6, Transition Mechanisms, Tunneling, 6to4 Tunnel, GRE Tunnel, Automatic IPv4-Compatible Tunnel, Manual Tunnel, 6PE, Dual Stack, Native IPv6, Performance Evaluation 1
Introduction The current version of Internet Protocol, IPv4 is widely used. It is easy to implement, robust, and supports a wide range of applications. IPv4 scales to the current needs for Internet connectivity. However, the growth of the Internet and address-hungry Internet services and applications has created deficiencies in IPv4. As more devices require connectivity to the Internet, IPv4 addresses will not be able to address this increased demand. IPv6 is the next-generation Internet Protocol that offers more IP addresses and overcomes the address exhaustion of IPv4. For latecomers to the Internet explosion, IPv6 is the only solution for them. Therefore IPv6 is expected to be widely used in the near future. MPLS is a label-swapping and packet-forwarding technology that is highly scalable and widely used by the service providers and large enterprises in the existing IPv4 backbone network. MPLS transfers packets by inspecting labels and forwards packets based solely on the contents of the label, rather than by performing complex routing lookups and examining the packets. The MPLS backbone is widely used to connect remote offices and headquarters to each other. The service providers and enterprises that have been using MPLS may view the integration of IPv6 services over an MPLS infrastructure as a normal evolution. The MPLS backbone can be used to connect islands of IPv6 with each other, either by using the existing IPv4 MPLS backbone or by partially or fully upgrading the MPLS backbone. In the event that the existing IPv4 MPLS backbone has to be used, there are multiple methods to provide connectivity to islands of IPv6. As the cost of fully or partially upgrading the backbone is high and requires the network to be upgraded, transition mechanisms are developed. Below are different methods evaluated in this paper that leverage existing IPv4 MPLS network and add IPv6 services without requiring changes to the backbone. These methods enable isolated IPv6 domains to communicate with each other over the existing IPv4 MPLS backbone. These approaches can be taken to avoid fully upgrading the MPLS backbone, resulting in minimal operational cost and risk. - IPv6 using tunnels between CE-to-CE routers, including Manual Tunnel, GRE Tunnel, Automatic IPv4-Compatible Tunnel, and 6to4 Tunnel - IPv6 using tunnels between PE-to-PE routers, including Manual Tunnel, GRE Tunnel, Automatic IPv4-Compatible Tunnel, and 6to4 Tunnel. - IPv6 on the PE routers (6PE) This paper also evaluates other methods of introducing IPv6 that require changes to the MPLS network. Dual Stack and Native IPv6 are the two methods evaluated in this paper that introduces higher operational cost due to upgrade of the MPLS backbone. Performance metrics, such as delay and jitter of the end-to-end system, and throughput of these methods will be analyzed in this paper and a statistical analysis will be performed. There are multiple objectives to be achieved in this research. 2
Objective 1 Perform a statistical analysis to determine if there is a statistically-significant difference between the performance metrics (delay, jitter, throughput) of IPv6 CEto-CE tunneling mechanisms, listed below. Then evaluate which method(s) is superior for each and overall performance metrics. Manual Tunnel GRE Tunnel Automatic IPv4-Compatible Tunnel 6to4 Tunnel Objective 2 Perform a statistical analysis to determine if there is a statistically-significant difference between the performance metrics (delay, jitter, throughput) of IPv6 PEto-PE tunneling mechanisms, listed below. Then evaluate which method(s) is superior for each and overall performance metrics. Manual Tunnel GRE Tunnel Automatic IPv4-Compatible Tunnel 6to4 Tunnel Objective 3 Perform a statistical analysis to determine if there is a statistically-significant difference between the performance metrics (delay, jitter, throughput) of IPv6 PEto-PE and IPv6 CE-to-CE tunneling mechanisms. Then evaluate which method(s) is superior for each and overall performance metrics. Objective 4 Perform a statistical analysis to determine if there is a statistically-significant difference between the performance metrics (delay, jitter, throughput) of the best tunneling mechanism(s), 6PE, Dual Stack, and Native IPv6. Then evaluate which method(s) is superior for each and overall performance metrics. IPv6 transition mechanisms are widely researched. Listed below are some of the researches that have been conducted on IPv6 transition mechanisms. The network performance of the configured and 6to4 tunnel on Windows and on Linux [9], [10] and the performance comparison of Teredo and ISATAP has been done [1]. The empirical performance of IPv6 versus IPv4 under Dual Stack has been done [8] and the Simulation of IPv4-to-IPv6 Dual Stack Transition between IPv4 hosts in an integrated IPv6/IPv4 network has been done [3]. The current deployment 3
and migration status of IPv6 has been done [4]. The IPv4/IPv6 Transition Mechanisms have been identified [11]. The IPv4/IPv6 Transition Technologies and Univer6 Architecture have been studied [2]. The simulation comparison methodology and framework for evaluating the performance of Novel IPv4/IPv6 Transition Mechanisms, specifically BD-SIIT versus DSTM has been done [6]. The examination of IPv4 and IPv6 networks, constraints, and various transition mechanisms has been done [5]. Although a great amount of research in this area has been done, to the best of our knowledge no one has performed a study on all the mechanisms identified in this paper over an MPLS network. Additionally, no research included the statistical analysis evaluating the performance impact over the long period of time with large data points. Moreover, this study is generating data with large number of applications such as Database Access, Email, File Transfer, File Print, Telnet, Video Conferencing over IP, Voice over IP, Web Browsing, and Remote Login. Finally this research has a systems engineering approach, by evaluating the performance of the end to end system including all the devices from the source to the destination such as clients, servers, switches, routers and the links connecting devices to each other with wide range of applications. Background This section provides an overview of the IPv6 transition mechanisms identified in this paper. Although there are other IPv6 transition mechanisms, such as Translation mechanism, they are not covered in this paper due to limitations of the OPNET simulation tool. 1) Dual Stack: In this transition mechanism, network devices such as routers and switches are configured with both IPv4 and IPv6 protocols. The Dual Stack devices use IPv6 stack to communicate with IPv6 devices and IPv4 stack to communicate with IPv4 devices. Although this is a good solution, it requires all the devices, including routers, switches, etc., to be upgraded to support both IPv4 and IPv6. This can obviously add to operational cost and risk of implementation. 2) Native IPv6: In this mechanism, all of the devices along the path from source to destination are IPv6-enabled only and there are no IPv4 devices or connectivity. This is the ultimate solution, but requires time and effort to get to this point. All commercial off-the-shelf (COTS) and homegrown applications should support IPv6 only. In the interim, other IPv6 transition mechanisms are more suitable, until the network, the devices and the applications are mature enough to support IPv6 only. 3) 6PE: In this transition mechanism, the MPLS core infrastructure is IPv6- unaware and only the PE routers are updated to support both Dual Stack and 6PE. The 6PE forwarding uses labels instead of IP headers. It has two labels; the inner label is limited to the advertised destination IPv6 prefix, and the outer label is related to the egress IPv4 address of the 6PE router. 4
The IPv6 reachability is between the 6PE devices using MultiProtocol - ibgp (MP-iBGP). The advantage of 6PE is that it does not require a tunnel, thereby avoiding the tunnel overhead. This is a very cost-effective solution for deploying IPv6 with minimal changes to the existing MPLS IPv4 network. 4) Tunneling: In this transition mechanism, one can carry IPv6 traffic using the existing IPv4 network by encapsulating IPv6 packets in the IPv4 header. At the tunnel end node, the packet is decapsulated and the IPv4 packet header is stripped. Then the original IPv6 packet is routed to its final IPv6 destination. The start and end nodes of the tunnel are IPv4/IPv6 Dual Stack-enabled and can be CE routers or PE routers. The tunnels are either manually configured or automatic. The main difference between the various tunneling mechanisms is the way that the source and destination of the tunnel are determined. Below are the different tunneling mechanisms used in this study. Manual Tunnel or Manually Configured Tunnel: This tunneling mechanism builds a permanent virtual link between two IPv6 networks that are connected over an IPv4 backbone. It is a point-to-point static tunnel. The start and end points of the tunnel have IPv4-routable addresses and an IPv6 address is configured on the tunnel interface. These tunnels are generally not scalable, because they have to be manually configured. They are used for stable connections that do not require much changing. GRE Tunnel or Manual GRE Tunnel: This tunneling mechanism is a type of manual tunnel with both tunnel source and destination configured manually for GRE. This tunnel has an extra encapsulation header for the GRE header. Therefore within the IPv4 backbone, the tunnel will have an IPv6 packet encapsulated into the GRE header and then into the IPv4 header. This tunnel is also used for stable connections that do not require much changing. Automatic IPv4-Compatible Tunnel: This tunneling mechanism has no preconfigured tunnels and the node performing the tunneling is assigned IPv4-compatible IPv6 addresses. The destination address is assigned automatically from the embedded IPv4 address of the IPv6 next-hop for the IPv6 route. 6to4 Tunnel: This tunneling mechanism is an automatic tunnel. In this tunneling, the destination is not explicitly configured and is obtained dynamically from the IPv4 address embedded in the destination IPv6 address of the packet. The IPv6 address of the tunnel interface starts with 2002: and the next 32 bits are the IPv4 address. This tunneling mechanism, unlike the manual tunnel, is not point-to-point and supports point-to-multipoint. 5
Experimental Setup The OPNET simulation tool SP Guru Network Planner 16.0 was used with the following specifications. - The CE routers were set as Cisco 3600 routers, the PE routers and the P (Provider) routers were set as Cisco 7200 routers, and the switches were set as Cisco 2940s. These routers were first configured in an emulated environment (GNS3) with IOS release 12.4(25) and detailed configuration below. Then the configurations for all the scenarios were imported to the OPNET Modeling and Simulation tool. - The MPLS cloud was configured with PE routers to have EBGP protocol for connectivity to CE routers and MP-iBGP protocol for connectivity to the remote PE router. The IGP routing protocol inside the MPLS cloud is Open Shortest Path First (OSPF). The appropriate redistributions were configured and the suitable address-family was configured for IPv4 and/or IPv6 depending on the transition mechanism. - The MPLS cloud was then configured to be IPv4-enabled and the hosts are IPv6 for the Tunneling transition mechanisms. If the hosts on each IPv6 island need to talk to each other at different islands they have to traverse the IPv4 MPLS cloud. The various tunneling mechanisms shown in Figure 1 were then configured with a total of eight tunneling scenarios. Four tunneling mechanisms - Manual Tunnel, Automatic IPv4-Compatible Tunnel, GRE Tunnel, and 6to4 Tunnel - were configured between CE-to- CE and the other four were configured between PE-to-PE routers. In case of CE-to-CE tunneling, the CEs were configured to be dual stack and the PEs and P routers were configured to be IPv4 only. In case of PE-to-PE tunneling, the CEs were configured to be IPv6 only, the PE routers were configured to be dual stack and the P router was configured for IPv4 only. 6
Figure 1. IPv6 CE-to-CE and PE-to-PE Tunneling Mechanisms Demonstrated in This Paper - The MPLS cloud is IPv4-enabled in case of 6PE with PE routers supporting dual stack, the CE routers supporting IPv6 and the P router supporting IPv4. For Native IPv6 the MPLS cloud and all the routers and switches were configured to be IPv6-enabled only. Finally for Dual Stack all devices were configured to be IPv4/IPv6-enabled. The hosts are IPv6 as shown in Figure 2. Figure 2. IPv6 Dual Stack, 6PE, and Native IPv6 Transition Mechanisms - The serial links between CE and PE routers are DS3, the links between PEs and P are Gigabit-Ethernet (GE) and the links to hosts are Fast- Ethernet (FE). The Maximum Transmission Unit (MTU) is set to 1500 bytes on all interfaces. 7
- The Windows hosts are Windows XP with receive buffer size of 16384 bytes. The Sun hosts are Solaris 2.9 with receive buffer size of 49152 bytes. The receive buffer sizes are the default settings for Windows XP and Solaris 2.9. - Traffic was generated using database access, email, File Transfer, File Print, Telnet, Video Conferencing over IP, Voice over IP, Web Browsing, and Remote Login applications. - The application profile was set such that the applications in each scenario would have the same transaction inter-arrival time, transaction size, and transaction mix (queries/total transactions). Also the variables such as transaction inter-arrival time, transaction size are set to be constant to ensure the model is consistent among all eleven scenarios. - The model was then ran with three seeds (128, 1103, and 21017), each 5 hours long for each of eleven scenarios. The results were set to be collected every second, resulting in 18,000 values per statistic for each seed. As a result, each scenario collected 54,000 values. The next step is to perform statistical analysis for evaluating the collected values and to accept or reject the hypothesis. Results The modeling and simulation tool OPNET was used and the applications have been set up as described in the Experimental Setup section above. The data has been collected for the eleven scenarios, each with 54,000 values. The next step for this research is to perform statistical analysis for delay, jitter, and throughput to identify if there is a statistically-significant difference between these scenarios and if so which one(s) are the superior methods. This paper will be updated to incorporate the results. References [1] Aazam Mohammad (2010) Comparison of IPv6 Tunneled Traffic of Teredo and ISATAP over Test-bed Setup. IEEE [2] Bi J, Wu J, Leng X (2007) IPv4/IPv6 Transition Technologies and Univer6 Architecture. IJCSNS International Journal of Computer Science and Network Security, Vol. 7, No. 1 [3] Chakraborty K, Dutta N, Biradar S (2009) Simulation of IPv4-to-IPv6 Dual Stack Transition Mechanism (DSTM) between IPv4 Hosts in Integrated IPv6/IPv4 Network. International Conference on Computers and Devices for Communications [4] Che X, Lewis D (2010) IPv6: Current Deployment and Migration Status. International Journal of Research and Reviews in Computer Science (IJRRCS), Vol. 1 No 2 [5] Govil J, Kaur N (2008) An Examination of IPv4 and IPv6 Networks: Constraints and Various Transition Mechanisms. IEEE [6] Hanumanthappa J, Manjaiah D, Aravinda C, (2010) An Innovative Simulation, Comparison Methodology and Framework for evaluating the Performance evaluation of a 8
Novel IPv4/IPv6 Transition Mechanisms: BD-SIIT vs. DSTM. First International Conference on Integrated Intelligent Computing [7] Huitema C (2006) Teredo: Tunneling IPv6 over UDP through Network Address Translation (NATs), RFC 4380 [8] Law Y, Lai M, Tan W, Lau Wing (2008) Empirical Performance of IPv6 vs. IPv4 under a Dual Stack Environment. IEEE Communications Society [9] Narayan S, Tauch S (2010) IPv4-v6 Configured Tunnel and 6to4 Transition Mechanisms Network Performance Evaluation on Linux Operating Systems. 2010 2 nd International Conference on Signal Processing Systems [10] Narayan S, Tauch S (2010) Network Performance Evaluation of IPv4/v6 Configured Tunnel and 6to4 Transition Mechanisms on Windows Server Operating Systems. 2010 International Conference On Computer Design And Applications [11] Punithavathani D, Sankarnarayanan K, (2009) IPv4/IPv6 Transition Mechanisms. European Journal of Scientific Research, Vol, 34 No.1 pp.110-124 [12] Sailan M, Patel H (2009) A comprehensive review of IPv4 and IPv6 for research test bed. International Conference on Electrical Engineering and Informatics [13] Shin M, Kim H, Santay D, Montgomery D (2006) An empirical analysis of IPv6 transition mechanisms. Proceedings of the 8 th International Conference on Advance Communications Technology (ICACT), Vol. 3, pp. 1990-1996 [14] Tahir H, Taa A, Nasir M (2006) Implementation of IPv4 Over IPv6 Using Dual Stack Transition Mechanism (DSTM) on 6iNet. Second International Conference on Information and Communication Technologies (ICTTA), Vol. 2, pp3156-316, ISBM:0-7803-9521-2 9