VoIP Performance over Different Interior Gateway Protocols

Transcription

1 VoIP Performance over Different Interior Gateway Protocols Xianhui Che, Lee J. Cobley Faculty of Applied Design and Engineering, Swansea Metropolitan University Mount Pleasant, Swansea, SA1 6ED, UK {xianhui.che; 34 Abstract: Voice over IP technology integrates data and voice networks and offers flexibility by supplying device interoperability using standards-based protocols. Routing is an essential data networking function that provides an efficient real-time data delivery VoIP requires. Best-effort networks leverage Interior Gateway Protocol technologies to determine paths for routing packets between hosts. Architects for best-effort networks provide services by over-provisioning the links and routers so that network congestion does not introduce unwanted levels of latency, jitter, and packet loss. Route re-convergence can be detrimental to VoIP users in the middle of a conversation, as it impacts both latency and jitter. This paper studies how VoIP performance can be affected by different routing behaviors which include Routing Information Protocol, Open Shortest Path First and Enhanced Interior Gateway Routing Protocol. Network modeling and simulation have been carried out with OPNET Modeler to evaluate and compare performances. Keywords: VoIP, Interior Gateway Protocols, RIPv1, OSPF, EIGRP, OPNET modeling and simulations. 1. Introduction VoIP technology offers cost-saving solution for integrated data and voice network, which will soon take off as a viable alternative to traditional voice systems and public switched telephone networks (PSTN) [1]. VoIP is flexible in driving new services by supplying device interoperability using standards-based protocols. VoIP as a real-time application brings new challenges for service providers and enterprises. Networks need to be more intelligent, secure, and have a higher level of performance. When designing a network to support VoIP and real-time applications, such considerations as application requirements, available budget, quality of service requirements, and downtime ramifications, have to be taken into account. Best effort network design is one of the common approaches to support VoIP service. Routing is an essential data networking function that provides an efficient real-time data delivery VoIP requires. Best-effort networks leverage Interior Gateway Protocol (IGP) technologies to determine paths for routing packets between hosts. IGP protocols are used within an autonomous system (AS) [2], which is described as a number of routers/network under single administrative control and share a common routing strategy. It is the largest entity within the internet hierarchy. Not all routing protocols understand AS, but the ones that do can control routes in and out of different autonomous systems, and set up network boundaries locking off different protocols or entire networks. IGP protocols build up routing tables which are referenced at each router hop traversed by the packets. Architects for best-effort networks provide services by over-provisioning the links and routers so that network congestion does not introduce unwanted levels of latency, jitter, and packet loss. Route reconvergence can be detrimental to VoIP users in the middle of a conversation, as it impacts both latency and jitter. The operations of routing protocols rely on two sets of information routing tables and distribution of knowledge [3]. While both are vital to a router s operation, different protocols will use different sets of knowledge and hold different information in the routing tables. But essentially they are the same. Routing tables are the mechanism that a router uses to locate other networks. These can hold information such as the next router that a packet has to be sent to, or hold more complex data such as information on every router in the local network and the speeds of the connected interfaces. The more complex a routing table becomes the more of the routers recourses such as RAM and CPU power become utilized. This can drastically slow down routers affecting the speed of the network. Distribution of knowledge is the way in which routers become aware of each others. The data sent, and the size of the data greatly varies between protocols; some send regular updates at periodic intervals, and others send information only when a change in the network occurs. This paper studies how VoIP performance can be affected by different routing behaviors which include the most widely used protocols such as Routing Information Protocol (RIP) and Open Shortest Path First (OSPF). Cisco s proprietary Enhanced Interior Gateway Routing Protocol (EIGRP) is also discussed for comparison purpose. Network modeling and simulation have been carried out with OPNET Modeler to evaluate and compare the performances. Section 2 outlines the design considerations for VoIP enterprise networks. Section 3 critically overviews the different interior gateway routing protocols. Section 4 explains the network models that are used for the comparisons, analyzes the simulation results, as well as evaluates the network performances. Section 5 concludes the paper. 2. VoIP in Enterprise Networks A private branch exchange (PBX) that provides VoIP service for customer sites can be either hosted or premisebased. Hosted solution mainly uses phones and some switches as on-site equipments. The switching and intelligence are remote. This type of network is cheap to run and easy to maintain, but the downside is that new services

2 35 are dependent on the provider, and the system is not flexible and cannot be customized [4]. The other solution is premisebased which is engaged with local switching and intelligence by the use of servers. It offers higher flexibility comparing to hosted solution, but the start-up cost can be high and extra complexity is introduced to the server maintenance and upgrades [5]. The main challenges facing the deployment of VoIP in large enterprise networks are the interoperability, security, and bandwidth management issues. These three problems, discussed below, are major stumbling blocks that keep VoIP technology from being implemented immediately into large corporations; until these problems are fixed, standard PBXs will remain merely for voice communications. 2.1 Interoperability The biggest challenge for network managers to overcome is the multi-vendor interoperability. Mainly there are a few important VoIP protocol stacks which have been defined by various standard bodies and vendors, namely H.323, Session Initiation Protocol (SIP) and Media Gateway Control Protocol (MGCP) [6]. While the ITU-T s recommendation H.323 is gaining wide recognition, many vendors have yet to completely comply with all the guidelines and other recommendations. Due to this reason, the technologies and equipments for implementing complete VoIP networks are still not operational at a feasible level. 2.2 Security Security has been a major concern in VoIP networks. Although H.323 defines encryption and authentication of user access, H.323-aware hackers can still tap into any conversation on the system, which means an employee or any outsider with internet access can monitor the voice conversations without ever having to leave the desk [7]. Another security issue arises if a corporation uses VoIP technology for a remote access location, which is one of the main uses for partial VoIP implementation today and often involves issues with firewalls. H.323 requires direct access to the company network and will open the entire network up to all UDP and TCP traffic [8]. The solution is to contain all H.323 traffic within one region and then use a voice trunk to connect traffic between the isolated region and the rest of the network. The other viable solution is to use an H.323-aware firewall [9]. 2.3 Bandwidth Management Another main challenge is the lack of bandwidth management of current networks employed within most large corporations. VoIP generates two types of network traffic the control messages, and the digitally encoded voice conversations. The control messages are used to setup and manage connections between IP phones and an IP PBX. The involved protocols normally use very little bandwidth and a delay of a few seconds in setting up a call is usually acceptable [10]. The real challenge is to satisfy the bandwidth demands of the digitized voice streams between users. Each conversation consumes a nearly constant amount of bandwidth for the duration of the call. The bandwidth required for each call depends primarily on the voice encoding technique as well as a couple of other variables [11]. Two voice encoding standards are widely supported by VoIP products G.711 and G.729. Again there is incompatibility problem with the codecs from different vendors. The bandwidth requirements for large companies or universities are much larger than for small companies and departments. Because the codec is the required components for converting analogue waves into packets of digital signals, the frame relays of large companies are over taxed [12]. Slower and more taxing working systems demand greater bandwidth for the codec. The reliance of large-scale implementation of VoIP technology is a big challenge for large corporations at the present time. 3. Reviews of Interior Gateway Protocols Dynamic routing protocols fall into three categories: Distance Vector (DV), Link State (LS) and hybrid protocols. The knowledge information shared by different network segments is defined by the routing protocol selected, which are stored in routing tables. To maintain an up-to-date routing table the router must determine the best information to be stored. Each protocol determines this based on certain criterion with the use of algorithms, which compile values known as metrics. Metrics are generated from as little as one characteristics of the network or more often several characteristics. The most common measurements normally include hop counts, delay, bandwidth, load, reliability (i.e. errors on the link), cost, etc. Among the three types of routing protocols, the simplest to configure is the distance vector protocol which use the distance and direction to find the best path to the destination by using an algorithm called the Bellman-Ford algorithm [13]. Network discovery is achieved by gathering information from directly connected neighboring routers which in turn may have gained their information from neighboring routers. To share this information distance vector protocols use a method known as a local broadcast. This sends out data to any device that is connected to an interface of the router. Distance vector does not care who receives and processes these broadcasts and that they are periodic in their approach. These protocols will be sent out updates at regular intervals regardless of whether or not there is a topology change. As these packets regularly traverse the network a large amount of unwanted network traffic, can be generated. Examples of distance vector protocols are RIPv1, IGRP, etc. Link State Protocols are slightly more complex than the distance vector protocols. These protocols use an algorithm called the Dijkstra algorithm [14] or shortest path first. This algorithm takes metrics into account when determining the best data delivery path. A typical example of link state protocol is OSPF. A major difference between distance vector and link state protocols is that whilst the distance vector protocol learns the routing table and distant routers from directly connected neighbors, link state protocols learn the entire topology of a network [15]. Link state protocol allows network devices to have a much better understanding and view of a network. Compared to distance vector protocols, link state protocols disperse information in a way which is much less bandwidth intensive. Link state protocols

3 36 use a method known as multicasting [16], which link state protocols uses to send updates when there is a change in the network and to specific hosts, whereas distance vector protocols periodically send out routing updates regardless of changes in a network. Hybrid protocols combine the mechanisms of both distance vector and link state protocols. The most common examples are RIPv2, EIGRP and Border Gateway Protocol (BGP). A typical mixture is a procedure that starts off with a distance vector protocol, and then adding more complex abilities of a link state protocol. Three protocols will be reviews in the following section: RIPv1, OSPF and EIGRP. 3.1 RIP Version 1 RIP version1 is a DV protocol that is easy to comprehend and deploy within an AS. Although superseded by more complex routing algorithms, RIP is still widely in smaller Ass thanks to its simplicity. RIP makes no formal distinction between networks and hosts. Routers typically provide a gateway for datagram to leave one network or AS and to be forwarded onward to another network. Routers therefore, have to make decisions if there is a choice of forwarding path on offer. The metric system RIP networks use is the hop count, which has a maximum value of 15 [17]. Every time a router passes the routing table to other routers a value of 1 is added to the metric inside the routing update. The maximum number of hop count is to solve the routing loops problem. Routing loops are basically confusions in a network topology that occur when the update/age out timers can be inefficient. With the hop count set to 15 the packet can be passed through a maximum of 15 routers before being discarded, without which the packets can be passed indefinitely until either the network crashes or the routers are switched off. RIP supports up to a maximum of 6 equal-cost path to a destination, this means that is a destination is reachable over different routes that have the same amount of hops, the router will hold all routes in memory up to a maximum of six (four is the default) [18]. The paths are all placed into the routers table and can be used to load balance when sending data. The main features of RIP can also lead to its disadvantages, such as information flooding, ineffectiveness of metrics system, and classful routing algorithm, explanations of which follow. Firstly, routing information is passed to other routers in a RIP network by using a local broadcast. This broadcast is by default every 30 seconds and is held for a maximum of 180 seconds [19]. The broadcast update contains the routers entire routing table; this is passed every 30 seconds among routers. These activities cause a fairly large amount of network traffic to be periodically sent throughout the network. This type of information flooding wastes network resources and cause network inefficiency and potential congestion problem. Secondly, the metric system that RIP uses is to find the shortest paths through a network for the data delivery. The task is carried out merely based on the hop count measurements regardless the other aspects of the networks such as bandwidth, etc. This behavior cannot guarantee the discovery of the optimal route for the data packets. Figure 1 shows an example. Given the network specifications, RIP would choose the route with the least number of hops rather than the optimal route which is also the fastest route. Should network congestion occur, RIP is able to balance the traffic load on different routes, however it can only take place on equal-cost paths. Figure 1. Ineffective route determination of RIP Thirdly, RIP comes under the heading of classful routing protocol; meaning only one subnet mask for any class of subnet can be used for the routers, which in effect can be wasteful of IP addresses. For instance, if is assigned to accommodate 6 subnets, subnet mask should be used, which prevents the use of the default mask of , otherwise, the router will return errors in the configuration file, e.g. duplicate IP Address. Non-meticulous subnet mask formality can cause loss of router configuration information which provokes unstable network performance. 3.2 OSPF OSPF is based on open standards and has good compatibility on a wider range of equipment, which is a prevalent routing protocol in larger enterprise networks. It is a LS routing protocol which uses more complex metric system to give efficient pathways discovery solutions to remote networks. The cost to measure the metric is worked out by taking the inverse of the bandwidth of links. Essentially a faster link is lower in cost. The lowest cost paths to remote networks are the most preferred routes, and held in the routing table. OSPF can load balance across a maximum of six equal-cost path links, although doing this can cause difficulties. The serial interface of the router is configured with a clock rate and a bandwidth. The clock rate is the speed data can be sent across a link, and the bandwidth is used by the routing protocol in the metric calculations. By default the speed of a serial interface is set to 1544 Kbps [20]. There is a potential hazard of this system. When different clock rates are set on a different link, the bandwidth has to be accordingly configured; otherwise OSPF will regard both connections as the same speed, which will cause problem with load balancing [21]. When routers need to run OSPF frequently, lots of resources are dedicated to the process; this potential problem can dramatically slow down the network service speed. There are some major differences between OSPF and RIP. Firstly, comparing to RIP, OSPF is a classless protocol

4 37 which allows utilization of different subnet masks, which essentially gives network administrators more flexibility with IP addresses and less wastage. Secondly, one appealing advantage that OSPF offers over RIP is scalability. OSPF has the knowledge of ASs and areas, and is able to understand the hierarchical routing structure. Thirdly, as a LS protocol, OSPF only sends out update information when there is a change in the network, rather than sending periodic updates at regular intervals as in DV protocols. This quality saves the bandwidth utilization throughout the entire network communications. Fourthly, while RIP uses broadcast to pass on routing information throughout networks which can cause potential network congestion problems, OSPF uses multicast method to reduce network traffic which uses addresses that are destined for particular machines. OSPF builds tables known as the adjacency and LS databases (LSD). The adjacency database is the database that holds a list of routers that the router has bi-directional connections with. The LS database lists all other routers in the network topology. Every router in a specific area will have the same LSD, which means every router has the same information about the state of the links and other routers neighbors. In case each router in a network needs adjacencies from one another, there would be a mass of information flowing through the network, plus the database of each router can be extremely large. The solution to this problem is the election of a router known as the Designated Router (DR) and Backup Designated Router (BDR) as shown in Figure 2, which are not necessary routers themselves but attributes of router interfaces [22]. DR is the central point of each area which forms adjacencies with all other routers within the area. The DR however could be a single point of failure which is regarded as a weakness in OSPF. This problem can be compensated by the designation of BDR which is a standby substitution for DR in case of failure. Figure 2. DR and BDR in OSPF 3.3 EIGRP EIGRP is one of the hybrid protocols, which is based on Interior Gateway Routing Protocol (IGRP). EIGRP has the ability to scale to an enterprise network size, not quite as large as an OSPF network can scale but a lot larger then a network running RIP can handle. EIGRP calculates distance by using a collaboration of different information. The characteristics selected are available bandwidth, delay, load, MTU and the link reliability. By using these factors the selected paths can be finely tuned, so information can be passed around a network by the faster most reliable routes. By default only bandwidth and delay are used. EIGRP is also a classless protocol and will support load balancing across six unequal paths [23]. This however is not such a simple command to use, and requires manual configuration. If incorrectly configured, it can cause network instability and routing loops, hence it is a common practice to ignore this ability. There are five components for the interworking of EIGRP protocol: neighbor tables, topology tables, route states, route tagging, and routing tables [24]. Neighbor tables are essentially a list of neighboring routers. There is a hold time that is set for each entry. If a router has not heard from a neighboring router within the specified hold time, then the router is considered non-operational, thus a failure recovery algorithm will be set in motion. Topology table holds all the destinations that are advertised by neighboring routers and the metrics linking them. Route states define the status of the routes that are held in the topology. Route tagging is an activity that identifies external routes among different ASs. Internal routes are then referred as the paths within ASs. Routing table contains information on all the routes that will be used to reach remote networks, such as advertised distance, and feasible distance, etc. EIGRP uses the Diffusing Update Algorithm (DUAL) to determine the routes. DUAL enables EIGRP routers to determine whether a path advertised by a neighbor is looped or loop-free. Low convergence delay can be achieved by maintaining a table of loop-free paths to every destination, in addition to the least-cost path. DUAL's convergence times are an order of magnitude lower than those of traditional DV algorithms [25]. The pseudo code for finite state machine of DUAL is: { Track all routes advertised by neighbors; Select loop-free path using a successor and remember any feasible successors; If successor is lost { Use feasible successor; If no feasible successor { Query neighbors and re-compute new successor; } } } A feasible successor route has to have the same metric as the successor route. In the example shown in Figure 3, if the

5 38 link between router 1 and router 2 fails, router 1 will use the feasible successor route to converge the network connection; hence network bandwidth is saved by the elimination of route re-computation. However, if the link between router 3 and router 4 breaks down, since router 3 has no feasible successors, it will query router 1 to obtain a new route. Figure 3. An example network of DUAL algorithm EIGRP is more flexible than OSPF. It has full support of distribute list [26]. Manual summarization can be done in any interface at any router within EIGRP networks. EIGRP offers fast network convergence and easy configuration. However, EIGRP is a Cisco-Proprietary protocol, which can affect the popularity of the protocol. Due to the expense of Cisco equipments not all networks will contain solely Cisco routers, and setting up some routers to use EIGRP and some to use alternate protocols can cause lots of confusion and waste of routers resources. 3.4 Summary of Comparisons Based on the above discussions, the comparisons of the three routing protocols are summarized in Table 1. RIP OSPF EIGRP NATURE DV LS Hybrid SCALE ROUTING METRICS DISCOVERY AND UPDATES FAILURE RECOVERY LOAD BALANCING Small networks Classful Routing loop counter mechanism Number of hops Periodical updates (broadcast) Slow convergence Only supported on equal-cost paths Enterprise networks Classless The inverse of the bandwidth of links DR multicasts whenever changes are made Generally faster than RIP [16] Supports 6 equal-cost paths, but difficult to implement Medium Classless 100% loop free Available bandwidth, delay, load, MTU and the link reliability DUAL Multicast Incremental update DUAL algorithm Supports 6 unequal paths, but commonly ignored due to its complexity and instability. Table 1. Comparison of RIP, OSPF and EIGRP 4. Performance Evaluation The VoIP performance metrics include delay, jitter, packet loss and Mean Opinion Score (MOS). Delay is the time that elapses between when an utterance is spoken and when it is played back at the receiver. Jitter is the variability in the delay which is computed as expected arrival time minus actual arrival time. De-jitter buffer helps fix the problem, but adds to the overall delay. According to ITU-T G.114 that recommends acceptable voice delay thresholds [27], the delay of VoIP network must be kept less than 150ms for real-time conversations and the voice jittering must be less than 30ms. Packet loss measures the percentage of the dropped packets which should be less than 1% [28]. ITU-T P standard defines MOS as a subjective metric which estimates the user satisfaction by means of a score which varies from 1.0 (poor) to 5.0 (best). The minimum MOS should be maintained at level 3 to achieve an acceptable performance [29]. Some of these parameters will be used to carry out the following performance evaluations. Table 2 shows a summary of the VoIP performance thresholds. Parameters Acceptable Level Delay 150ms Jitter 30ms Packet Loss 1% MOS 3 Table 2.Threshold of acceptable VoIP performance This study uses OPNET Modeler as the simulation tool [30]. It is a commercial network simulation software package which provides a platform for modeling and simulating network applications. An enterprise-scale network has been built with OPNET as shown in Figure 4. A few routers (R1, R2, R3, R4 and R5) are connecting two office branch networks that have 100BaseT specifications. Various bandwidths have been configured for the links interconnecting the routers in order to create different routing metrics parameters. Two bottleneck channels have been generated: 56K data rate between R1 and R2, and 33k data rate between R2 and R5. Three network scenarios have been designed to enable the three routing protocols respectively. Voice traffic is running through the network, which has the identical application configurations and user profiles for the three scenarios. The simulations are scheduled to run for half an hour. In order to evaluate the network performance in reaction to network failures, the link between R1 and R3 are deliberately failed after 10 minutes, and after another 10 minutes it is set to be automatically recovered by OPNET. RIP versus OSPF/EIGRP Figure 5 shows the total voice traffic received by the end users. Figure 6 displays the number of hops per route. Figure 7 indicates the MOS values. Figure 8 demonstrates the voice jittering situation. As expected, RIP chooses the route that has the least number of hops (Figure 6), despite the existence of a bottleneck transmission. The inefficient data delivery leads to poor throughput (Figure 5) and poor MOS value (Figure 7). When the network is initializing, routing traffic occupies network bandwidth that causes congestion problem in the bottleneck links, which causes serious jittering in the

6 39 start of the voice conversation (Figure 8). In contrast, before the failure point emerges, EIGRP and OSPF have similar performances, i.e. high-throughput (Figure 5), acceptable MOS values (Figure 7), and few instances of jittering (Figure 8). The reason for that is they both take link bandwidth into account when choosing the optimal routes; hence faster routes will be chosen to accommodate high profile traffic. Figure 6 indicates that it is likely that EIGRP and OSPF have chosen the same route for the VoIP application before the failure point. EIGRP on the other hand, seems to be dysfunctional during the link failure state, as non statistics are collected in Figures 7 and 8. Figure 6 implies that the data delivery process proceeds for one hop and stops at R1. The conjecture is no feasible successor could be found by the DUAL algorithm; hence the network is going through slow convergence period to re-negotiate the route which involves queries of neighbor s routing tables, and re-computing of bandwidth. While the re-convergence progress is still ongoing, the failed link is restored by OPNET, so EIGRP is then re-engaged with the original routing process. The results indicate the inefficient network re-convergence behavior of EIGRP when there is a failure point and no immediate feasible successors are found. Figure 4. Network simulation model With the introduction of network failure point, EIGRP and OSPF start to yield different performances. At this point, it is worth mentioning that RIP is not disturbed by the failed link in the simulated network as shown in the statistical results, because the failed link is not part of its chosen route. Based on the anticipation of slow convergence conduct of RIP, the objective of this part of the simulation is to compare the network failure recovery behavior of EIGRP and OSPF. OSPF versus EIGRP The link failure has affected the performance of both EIGRP and OSPF. During the deliberate link failure and auto-recovery process, OSPF acts consistently throughout the procedure, while EIGRP is heavily disrupted during the failure but restores to the original state after the failure recovery. Figure 6 shows that upon the failure point, OSPF has chosen an alternative route to continue the date delivery process, and sticks to that route even after the failure is recovered. As explained in the previous sections, OSPF only updates the routing table whenever changes are made. Simulation results imply that it will not alter the route for any existing traffic stream as long as there are no congestions or other problems in its chosen route. In the network model shown in Figure 4, all the possible substitute routes (i.e. linked by R3 and R4) have rather high bandwidths. So even though the throughput is reduced after the network reconvergence (Figure 5), the MOS and the jittering are all maintained in the acceptable level (see Figures 7 and 8), which indicates the flexibility and efficiency of OSPF for the VoIP service network. Figure 5. Voice traffic received (bytes/sec) Figure 6. Number of hops per route Figure 7. MOS value

7 40 5. Conclusions Figure 8. Voice jittering Although VoIP offers great benefits for service providers and enterprises, challenges to implement VoIP application over enterprise network still remain. Routing is an essential data networking function that provides an efficient real-time data delivery required by VoIP. Best-effort networks leverage IGP technologies to determine paths for routing packets between hosts. Three IGP protocols are discussed RIP, OSPF and EIGRP, of which the features as well as the advantages and disadvantages have been analyzed in the paper. An enterprise-scale network has been built with OPNET Modeler. Various aspects of network have been specified in the simulation model in order to evaluate the performance of the three routing protocols. As expected, RIP carries out low-efficient routing in the network with a bottleneck transmission link as it does not take bandwidth into consideration. In contrast, RIP and EIGRP perform with excellence as they are devoted to computing the fastest possible route. Results show that with the same network specifications it is likely that EIGRP and OSPF have chosen the same route for the VoIP application. The link failure has affected the performance of both EIGRP and OSPF. During the deliberate link failure and autorecovery process, OSPF acts consistently throughout the procedure, while EIGRP is seriously disrupted during the failure but restores to the original state after the failure recovery. OSPF updates the routing table upon network failure to re-calculate a new route, and does not alter the route for any existing traffic stream as long as there are no congestions or other new problems in its chosen route. Network based on OSPF maintains acceptable performance throughout the process, which indicates its flexibility and efficiency for the VoIP service network. On the other hand, the DUAL algorithm is not as efficient as OSPF when no feasible successor is found. The statistical analysis has led to the conclusion of resilience and efficiency of using OSPF in enterprise networks to support VoIP service. References [1] J. Slay, M. Simon. Voice over IP forensics. In Proceedings of the 1st international conference on Forensic Applications and Techniques in Telecommunications, Information, and Multimedia, and Workshop, article no.10, [2] B. Fortz, J. Rexford, M. Thorup. Traffic engineering with traditional IP routing protocols. IEEE Communications Magazine, pp , vol. 40, Oct [3] C. Metz. IP routers: new tool for gigabit networking. IEEE Internet Computing Magazine, pp , vol. 2, Nov-Dec [4] S. Hanson, VoIP in the Enterprise, Tiscali Technical Report, Mar [5] Introducing VoIP to the Enterprise Using an IP PBX. Intel Application Report, [6] B.Chatras, S. Garcin. Service drivers for selecting VoIP protocols. In Proceedings of 11th International Telecommunications Network Strategy and Planning Symposium, June [7] W. Marshall, A.F.Faryar, K.Kealy, G. Reyes, I. Rosencrantz, R. Rosencrantz, C. Spielman, Carrier VoIP Security Architecture, In Proceedings of 12th International Telecommunications Network Strategy and Planning Symposium, [8] G. Goth. VoIP Security Gets More Visible. IEEE Internet Computing, pp. 8-10, vol. 10, Dec [9] C. Feng, S. Malik. Vulnerability analysis and best practices for adopting IP telephony in critical infrastructure sectors, IEEE Communications Magazine, pp , vol. 44, Apr [10] A. P. Markopoulou, F. A. Tobagi, M. J. Karam. Assessing the quality of voice communications over Internet backbones, IEEE/ACM Transactions on Networking, pp , vol. 11, Oct [11] L. Ding, A. Radwan, M. S. El-Hennawey, R. A. Goubran. Performance Study of Objective Voice Quality Measures in VoIP. In Proceedings of the 12th IEEE Symposium on Computers and Communications, [12] M. C. Schlesener, V. S. Frost. Performance evaluation of telephony routing over IP (TRIP). In Proceedings of the 3rd IEEE Workshop on IP Operations and Management, pp , Oct [13] G. Cheng, N. Ansari. Multiple additively constrained path selection, IEE Communications, pp , vol. 149, Oct-Dec [14] Y. Fujita, Y. Nakamura, Z. Shiller. Dual Dijkstra Search for paths with different topologies. In Proceedings of IEEE International Conference on Robotics and Automation, vol. 3, Sept [15] J. Zeng, F. Chen, G. Wei. Practical traffic engineering based on traditional IP routing. In Proceedings of International Conference on Wireless Communications, Networking and Mobile Computing, pp , vol. 2, Sept [16] H. Pun. Convergence Behavior of RIP and OSPF Network Protocols, E-book resources, Communication Networks Laboratory, Simon Fraser University.

8 41 [17] J. Wang, F. Xu, F. Sun. Benchmarkinng of Routing Protocols for Layered Satellite Networks. In Proceedings of Multiconference on Computational Engineering in Systems Applications, pp , vol. 2, Oct [18] M. Sportack. IP Routing Fundamentals, Cisco Press, Feb [19] P. Almquist, F. Kastenholz. Towards Requirements for IP Routers, RFC 1716, [20] J. Doyle, J. Carroll. Routing TCP/IP. Cisco Press, Oct [21] J. Wang. Load Balancing in Hop-by-Hop Routing with and without Traffic Splitting. University of Illinois at Urbana-Champaign, Jan [22] D. Sidhu, T. Fu, S. Abdallah, R. Nair, R. Coltun. Open Shortest Path First (OSPF) Routing Potocol Simulation. In Proceedings of SIGCOMM, Oct [23] B. Augustin, T. Friedman, R. Teixeira. Measuring Load-Balanced Paths in the Internet. In Proceedings of the 7th ACM SIGCOMM Conference on Internet Measurement, Oct [24] T. M. Jaafar, G. F. Riley, D. Reddy, D. Blair. Simulation-Based Routing Protocol Performance Analysis: a Case Study. In Proceedings of the 38th Conference on Winter Simulation, Dec [25] S. McQuerry, Interconnecting Cisco Network Devices, Cisco Press, [26] White paper. EIGRP to OSPF Migration Strategies. Juniper Networks, [27] One-Way Transmission Time, ITU-T Recommendation G.114, May [28] J. Chapman, "Quality of Service for VoIP," Cisco Systems, [29] Mean Opinion Score (MOS) terminology, ITU-T Recommendation P.800.1, Jul [30] OPNET Modeler,