D uring the last decades people have become witnesses to the rapid expansion of data networks. New ways of working, communicating and socializing have been accompanied by new terms such as “Bring your own device” (BYOD), the “Internet of things” and of course Cloud services. Routing protocol evolution has followed a similar path, simple distance-vector protocols and default routes led the way to link-state and hybrid protocols. This evolution is attributed to the following business demands: fast convergence in data centers in order to accommodate redundancy and failover requirements, public and private clouds service level management, and higher percentage availability. The focus of this thesis is based on the two internal gateway protocols, OSPF and EIGRP, which monopolize the market, as more or less industry standards. Both operate inside an autonomous system, and while having different ways to form relationships among routing devices, adapt to topology changes and handle failures, they provide the same results: robust routing tables and network stability. The objective is to evaluate their performance characteristics and more specifically their network convergence duration, video conference packet delay variation, IP voice jitter and CPU utilization in two different topologies. The first topology provides the basis for simulation comparison, while the second examines the scaling effects of both protocols.
Keywords: OSPF, EIGRP, network convergence duration, packet delay variation, IP voice jitter, CPU utilization, throughput, dynamic routing protocols, OPNET.
I n the procedure of network communication, routing is the process through which routers move information across networks, while they become aware of where to forward information in order to arrive at its final destination. Since networks have become extremely large, Organizations, Enterprises and ISPs rely heavily on routing protocols, both internal and external. These protocols assist routers in finding neighbors, keeping track of neighbor relationships, learning new routers and recover quickly from sudden failures of connected or remote links. This makes evident that a poor choice of a routing protocol can lead to dismissed performance, routing-loops, and low quality of service even in small and medium-sized networks. To choose an efficient routing protocol, one must first decide whether this will be used inside an autonomous system that is in a network under a common administration team, or between different autonomous systems. In such case, an external routing protocol should be utilized. Furthermore, other factors that must be taken into account are the size of the network, the hierarchical structure, if one exists, multiple equal or unequal paths to networks, and bandwidth of the links. These lead to different consideration and decisions affecting factors such as convergence, meaning the speed of routing tables build-up in response to changes in the topology, the amount of traffic the protocol send in order to service its needs and of course the CPU and memory utilization that imposes on the router.
T he aim of this thesis is to compare different performance characteristics of the OSPF and EIGRP routing protocols for IPv4. This will be accomplished by attending to the following objectives:
Two topologies will be created in a simulation environment. A simple one which will provide the control basis of examining the default characteristics and behavior of each protocol in question. A more complex topology, depending on the capabilities of the simulation tool, will demonstrate the scaling effects of each protocol.
Investigation of the default behavior of each protocol in non-hierarchical metropolitan topologies.
Examination of the different timer strategies that can be implemented to enhance the routing process, especially the network convergence duration and the packet delay variation outcomes.
Measurement of network convergence duration, video conference packet delay variation, IP voice jitter and CPU utilization in all aforementioned scenarios.
The comparison between the two protocols will be performed in two stages. Through the analysis of their features and operation as well as through various simulation experiments. Therefore this thesis is divided in two parts:
In the first part the OSPF and EIGRP protocols are presented, analyzed and explained carefully based on their operational and convergence behavior. Since they implement unique algorithms, namely SPF and DUAL, and use different metrics based on cost (OSPF) and bandwidth, reliability, load and reliability (EIGRP), they operate differently on topologies which when scaled present non-hierarchical shapes or non- efficient route summarization structures.
The impact of the inherent behavior of each protocol will straight forwardly affect the performance in such cases which will be demonstrated through this simulation experiments. Although it is a fact that careful configuration and usage of the EIGRP will provide better results, this thesis will attempt to present that this may not be true especially on network convergence duration, if the large network is not designed with a hierarchical structure. In this case, the query process that will be initiated when a link suddenly fails, is expected to have a great impact on the convergence time and end-to-end delay of traffic in the network.
i.Convergence of the OSPF protocol will occur faster in simpler topologies compared to the convergence of the EIGRP protocol.
ii.Convergence of the EIGRP protocol will occur slower in non-hierarchical scaled topologies than that exhibited by the OSPF protocol.
Packet Delay Variation of the EIGRP protocol will be the same or slightly lower than that of the OSPF protocol leading to higher throughput.
The IP Voice Jitter of the EIGRP protocol will be lower than that of the OSPF Protocol.
i.OSPF will require more computational resources in simpler topologies compared to EIGRP.
ii.EIGRP will demand more processing power in non-hierarchical scaled topologies compared to OSPF.
T This thesis was conducted by assessing available technical and scientific literature mostly from IEEE & CISCO as well as through the University library. Additional relevant data was reviewed using a range of information sources such as numerous books, bibliographic databases, conference proceedings, networking journals and technical publications not to mention internet search engines in order to explore the EIGRP and OSPF comparison from different perspectives. To aid the searching technique, specific key words were used with terms related to network convergence duration, throughput, packet delay variation, jitter, CPU utilization, dynamic routing protocols, simulation and OPNET for the proper completion of the research. A basic notice, however, was that limited literature is available on the educational edition of the Riverbed Modeler Academic Edition 17.5 simulation tool. Following the completion of the literature review and the extended research of the characteristics of EIGRP and OSPF, scenarios designed and defined the quantitative metrics by which the performance of the two dynamic routing protocols was compared. Proceeding, through the infrastructure set up and the run of the scenarios on the simulation tool the results collected in order to analyze and determine if the hypothesis was verified, namely that EIGRP has better performance than OSPF.
I n the first main part of the research the two protocols that will be examined are presented. Firstly the OSPF, which is an open-source protocol, an industry-standard and implemented in routers produced by all network equipment companies. OSPF is an older dynamic routing protocol based on the Shortest-Path-First algorithm developed by Dijkstra. The SPF is an algorithm with optimal performance but imposes a CPU and memory burden on the router. Afterwards the EIGRP, which is a proprietary Cisco protocol, a de-facto standard as well as a newer protocol designed by J.J.Garcia-Luna-Aceves (Cisco, 2015). This dynamic routing protocol takes into account more parameters in order to form its composite metric. It also avoids routing loops using an efficient assumption called “feasibility condition” which populates a topology table containing all backup routes to each destination, thus making the protocol more efficient in sudden network failures. In the second main part of this thesis first a simple topology will be designed in order to provide the basis of the control experimentation. Both protocols in situations of varying load will be tested as well as their performance in small networks will be examined. Furthermore, a scaled-up topology will be designed, which will allow OSPF to assume an advantage since it will implement by default a hierarchical approach to routing. EIGRP is expected to have a more difficult job since random network failures will push it to work more intensively as well as the network will not provide hierarchical subnetting or allow efficient automatic summarization.
R outing protocols have evolved continuously since the invention of the Internet. Starting from RIPv1, RIPv2 and IGRP which demonstrated significant limitations in classless networks and discontiguous topologies, different approaches emerged (Fortz et al., 2002). The first approach necessitated the usage of optimal algorithms to decide on the best network routes. These algorithms should entail characteristics such as speed and efficiency to arrive at loop-free network representations. This formed the basis for the development of OSPF, and its hierarchical structure. It must be reminded here that multi-area OSPF implements a backbone, to which all other areas must connect. This provides stability in the networks and with proper design does not allow instabilities to be transferred across areas (Goyal et al., 2012). Although OSPF is a complete protocol, meaning that it takes into consideration all links in a network and carries knowledge of the whole topology, it needs to do extensive recalculation in the case of network failures. This led to the development of a distance-vector protocol which imitates the loop-free decision made by OSPF by an assumption. This assumption, known as the “feasibility condition”, tested through simulation experiments and according to Cisco (2015) enforces each router to test the information it receives from its neighbors.
Several studies have investigated the behavior of these protocols, both separately and in comparison, demonstrating how traffic is affected by different routing behaviors (Al-Saud et al., 2008; Kaur & Sharma, 2011). Factors such as convergence and delay due to link failures have been analyzed leading to the experimental conclusion that EIGRP provides better results, while other studies show mixed results (Thorenoor, 2010; Wijaya, 2011; Wu, 2011). This is the reason that leads the research in this thesis: to examine what is the behavior of EIGRP and OSPF in a scaled-up topology with no automatic summarization and not careful hierarchical design on the part of the network architect. This may be the case in networks that are created through the merging of different companies, thus the merging of different structured autonomous systems. This will be better evaluated in situations of heavy load and sudden network failures. This will impose further burdens on the convergence processes and end-to-end network delays.
According to earlier surveys, the performance of EIGP and OSPF dynamic routing protocols was compared based on various parameters. Ittiphon & Suwat (2012) claim that EIGRP is more efficient than OSPF concerning the rerouting and the retransmission time required to reach the network destination in link-failure circumstances. Moreover, Shafiul et al. (2013) extended their research on the comparative performance analysis of EIGRP & OSPF dynamic routing protocols through real time video streaming applications. It must be noted that the evaluation of the results according to packet loss, throughput, end-to-end delay and convergence duration indicated that EIGRP provides a better performance than OSPF for real time applications. Krishnan & Shobha (2013) executed the exact same research through the simulated network model which lead them to the fact that EIGRP performs better compared to OSPF not to mention that less system resources are needed from EIGRP thus lesser heat is produced in opposition to OSPF. Thorenoor (2010) conducted an approximate research which involved the aforementioned quantitative metrics as well as the bandwidth and CPU memory usage. The assessment results exhibit that EIGRP requires less bandwidth and CPU memory utilization providing better network convergence time compared to OSPF.
Continuing the extensive research on the general performance of the two routing protocols comparison in this thesis, it is observed that Jun, Xiaoxiang & Jianping (2006) after a thorough examination of OSPF regarding the network scalability concludes that this Open Shortest Path First protocol faces problems with the CPU usage. Che, & Cobley (2009) examined the VOIP performance over RIP, EIGRP and OSPF using the same simulation tool as in this thesis. As expected RIP couldn’t reciprocate on the VOIP routing requirements in comparison to EIGRP and OSPF. However EIGRP in link failure situations is not as effective as OSPF with the absence of feasible successor. OSPF maintains efficient and flexible performance throughout the process re-calculating a new route without altering “the route for any existing traffic stream as long as there are no congestions or other new problems in its chosen route”. Last but not least, Kalyan & Prasad (2012) declare that many factors play a crucial role upon the selection of the protocol that will be used in each circumstances. They insist that a single protocol is not able to cover all the prerequisites at all times. That is why comparisons in both dynamic protocols like in this thesis are conducted in order to facilitate the decision of which protocol performs better over four parameters covering medium to large networks.
Multiple theories emerge regarding the performance of the two protocols as all these analysts and scientists report the results that have come in over the years. However, as the technology evolves and new versions of simulation tools with more capabilities are created, the comparison that will be conducted in this thesis will be more accurate. The reason this research should be discussed further is that while EIGRP is considered a better protocol, it presents some disadvantages in comparison to OSPF. While the hypothesis of this thesis that EIGRP performs better than OPSF, through the extensive Literature Review and Computer-Based Simulation comparison between EIGRP and OSPF, this thesis will present and analyze the characteristics of these two dynamic routing protocols as well as it will evaluate the performance results of the simulation tool based on various parameters.
I In order to form a network, in most cases people start by connecting the endpoint devices (PCs, tablets, smartphones, servers, printers etc.) through their network interface cards (NICs) to switches or wireless access-points. In this fashion Local Area Networks are created, which operate mainly at the Data-Link Layer (Layer 2) of the OSI networking model (Sendra, et al., 2011). Yet, how different local area networks connect? The solution is to implement routing. Routing is the process of selecting a path to a destination and is performed by routing devices (routers, Layer 3 switches or servers) at the Network Layer (Layer 3) of the OSI networking model. The packets are examined and they are “routed” to their destination by taking into account the Layer 3 destination address. Although routing has been performed initially by servers, later on it was assigned specialized equipment namely routers (Doyle, 2001). And how routers operate? First of all, routers are connected to multiple networks. When they receive a packet on one of their interfaces they examine whether the packet is destined to the same network this interface belongs to. In this case, they simply ignore the packet. But if the packet is destined for a different network, then they perform a lookup operation, searching through their routing table, in other words a local database, to find an exit interface to forward the packet. Therefore a router performs two operations, a lookup process to find a route in their routing table, and a switch operation to take a packet from one interface and encapsulate it again to be sent to a different interface (Kurose & Ross 2004).
The most crucial part: how are routing tables created? At first, the router inserts to the routing table all the different networks that are directly attached to it and are operational. Then it inserts all networks that are configured by the administrator through static route commands. Finally, if a dynamic protocol is configured and running, the router inserts all routes learned through this protocol into the routing table. If all above steps are performed, then the routing table is dynamic and changes whenever there is an update in the network topology (Medhi & Ramasamy, 2007). Dynamic protocols are divided in different categories based on whether they operate inside or outside an autonomous system (interior or exterior gateway protocols), or on whether they implement a distance-vector or link-state protocol. An autonomous system is a set of routers that operate under the same administrative control and can encompass a very large number of routing devices. Examples of interior gateway protocols are: RIPv1 and RIPv2, IGRP, EIGRP, OSPF and IS-IS while the industry standard in exterior gateway protocols is the BGP (Medhi & Ramasamy, 2007; Kurose & Ross, 2004).
By the term distance-vector routing is meant that routing decisions are made based on vectors of routes (along with the corresponding distances) learned by directly connected neighboring routing devices. It is a fact that routers that implement distance-vector routing do not know the entire network topology but only have knowledge of the distance from the destination network and the direction that traffic must be forwarded. Routing protocols that belong to the distance vector category are: RIPv1m RIPv2, IGRP and EIGRP (Xu, Dai & Garcia-Luna-Aceves, 1997). By the term direction is meant that a route is discovered by the interface of a router and by the term distance is meant the “cost” to reach a network destination. This “cost” can be measured in “hops” (routing devices) in the case of the RIP protocol or through a composite metric in the case of IGRP and EIGRP, taking into account factors such as bandwidth, delay, load and reliability (Vutukury & Garcia-Luna-Aceves, 2001). One of the main characteristics of distance-vector routing is that updates are sent periodically to all interfaces. These updates may contain the whole routing table or a part of it (partial updates). When a participating router receives such an update, it compares to what it already knows in its routing table, encompasses all new information, renews existing information and then floods what it knows to its neighbors (Pei et al., 2004). This type of routing has some inherent problems concerning the creation of routing loops, in the case that multiple paths exist to a destination. This happens since distance-vector routing is called routing-by-rumor. Each router does not have an explicit idea about the whole network topology but believes what its neighbors are telling him. Various ways have been developed to deal with this problem and these can be summarized to: counting to infinity, split-horizon and poison reverse. Since these are not a topic of this thesis they are not further analyzed (Rakheja et al., 2012).
By the term link-state routing is meant that routing decisions are made individually on each router based on a network graph that exists in its memory. This graph contains the connections of all nodes on the autonomous system (all existing operational links). This topology information permits each router to calculate the best path or paths to all different networks in a system, which are then placed on the routing table. A main characteristic of this process is that a router needs not to update periodically its neighbors but only when an event occurs (new router discovered or sudden link failure). Routing protocols that belong to the link-state category are: OSPF and IS-IS (Liu & Reddy, 2004). Link-state routing starts with the neighbor discovery phase where every router exchanges hello packets to find the neighbors on all operational links and maintain relationships with them. Afterwards, each router floods its connected links, so that all the routers inside the autonomous system learn the links and those who are producing these. All these links end in a link topology table maintained by each router. This table along with the neighbor table allows each router to form a complete topological view of the network (Haas & Pearlman, 2001).
The final stage is the execution of an algorithm which produces the shortest path to each link on the network, based on the link cost parameter. A network graph is created and the router starts executing a shortest path algorithm by putting itself as the root of the output tree. The final output of the algorithm, which runs independently on each router, populates the routing tables inside the autonomous system. A characteristic feature of the algorithm is that alterations in the topology lead to the re-computation of the shortest path algorithm and as a result to a CPU and memory burden (Hinds, Atojoko & Zhu, 2013). This type of routing has an inherent advantage over distance-vector routing. Since all routers have knowledge of the whole topology, and in particular the same view of the network topology, the formation of routing loops is more difficult to happen.
Open Shortest Path First is an Interior Gateway Routing Protocol for Internet Protocol (IP) networks and its research originated from a working group of the Internet Engineering Task Force (IETF) as early as the 1970s, with some implementations on the Arpanet. OSPF belongs to the link state routing protocol family and is used in order to distribute routing information within a single Autonomous System. It must be noted that the name of this protocol depicts its two main characteristics. The first word Open refers to the fact that the protocol was developed using the open and public RFC (Request for Comments) process and the SPF (Shortest Path First) refers to the well-known algorithm by Dijkstra which dynamically determines the shortest path through a network. In 1989, the first OSPF version was created (OSPFv1) and drafted in RFC 1131. In 1991, the second version (OSPFv2) was drafted and revised in RFC 1583, 2178 and 2328. Finally, in 1997 the OSPFv3 for IPv6 was released in RFC 2740 (Moy, 1998; Ferguson & Moy, 2008).
In general, an OSPF message is encapsulated in a packet as followed (Graziani & Jonson, 2008):
More specifically, the OSPF Packet Header is included with every OSPF packet and is encapsulated in an IP packet with a Protocol field of 89 while the destination address is either the multicast address 184.108.40.206 or the 220.127.116.11. The OSPF Packet Header is depicted in the table below (Graziani & Jonson, 2008; Ferguson & Moy, 2008):
Furthermore, OSPF uses five different packet types. Each one serves a different purpose (Ferguson & Moy, 2008). These are:
1. Hello Packet 2. DataBase Description (DBD) 3. Link State Request (LSR) 4. List State Update (LSU) 5. List State Acknowledgement (LSAck)
a. Hello Packet (OSPF Type 1 packet) is used to discover the neighbors and exchange the routing databases. Through this packet, certain parameters are advertised and when their match is accomplished the adjacent router becomes neighbor. Furthermore, Hello packets are utilized as a keep-alive mechanism. Specifically, Hello packets are sent periodically to their neighbors in order to obtain bidirectional communication. If a router does not receive Hello packets from its neighbor in a particular interval (Dead interval), then the neighbor declared down (presumed dead) and all the information acquired through it is invalidated. It must be stressed that on Broadcast or NBMA networks, the Designated (DR) and Backup Designated Routers (BDR) routers are being elected through Hello packets (Graziani & Jonson, 2008). b. Database Description Packets (OSPF Type 2 packets) are exchanged when an adjacency is formed providing the link state Database topological content. The receiving router verifies the local link state Database through a poll-response procedure between master and slave routers. c. Link State Request Packets (OSPF Type 3 packets) are used to request more information about the topological Database after the exchange of the Database Description packets with the neighboring routers. This constitutes the last stage for the creation of the adjacency. d. List State Update Packets (OSPF Type 4 packets) purpose is to flood the link state advertisement. Several Link State Updates are included in a single Link State Update. e. List State Acknowledgement Packet (OSPF Type 5 packet) are sent and received in order to make the flooding of the multiple link state advertisements via the LSU packets reliable.
Router LSAs (Type 1) are generated by all routers for each area that belong to. These are only flooded within a particular area and in no case they cross areas. Network LSAs (Type 2) are generated by the Designated Routers and describe all routers that are connected in a specific segment of the network. They are flooded only within the area. Summary LSAs (Type 3 and 4) are generated by Area Border Routers (ABRs) in order to advertise inter-area routes to the other areas in an Autonomous System. Type 3 messages (summary links) aggregate routes between areas, while Type 4 messages describe routers through which the ASBR can be reached. With Type 4 messages all routers are aware of routers that lead outside the Autonomous System. External LSAs (Type 5) are generated by ASBRs in order to inform all routers on external routes to the Autonomous System. These routes are redistributed in OSPF and are flooded all over except the stub areas.
The OSPF protocol uses as its metric the cost of an interface. This is inversely proportional to the bandwidth of the interface. It is profound that the higher the bandwidth of an interface the lower the cost.
The cost of an interface in OSPF routing protocol is determined by the formula above. It should be mentioned that the value of 108 is equal to 100.000.000 in bps and the cost of an interface is measured by default based on the bandwidth. Moreover, the cost would be easily attained if the reference bandwidth is divided by the interface bandwidth (Cisco, 2005).
O SPF is a link state routing protocol which implements the shortest path first algorithm to determine the path with the least costs to all known destinations. The shortest path to all destinations are calculated using the Dijkstra algorithm which provides an optimal solution considerably convoluted. Several processes of the algorithm are detailed below:
A link-state advertisement is generated by the router whenever a change in an attached network occurs or during initialization.
LSAs are being exchanged through the flooding procedure between all routers. Each router stores the identical link-state update that has received, in its link-state database and afterwards propagates the link-state update to other routers.
When the creation of the link-state database in every router is accomplished, the router running the Dikjstra algorithm, creates a shortest path tree to all destinations.
If something changes in the network, such as a flapping interface or a change in link costs, the link-state protocol propagates these throughout the network, allowing all routers to keep-up to date their topological information (Graziani & Jonson, 2008; Ferguson & Moy, 2008).
In order for each router to create its routing table, it utilizes the neighbor table, the topological information and the shortest-path first algorithm. It assumes that itself is the starting point and calculates a loop-free topology by running the SFP algorithm, examining in turn all the topological information learned by adjacent routers. In the following figure is depicted how a physical topology is transformed to a tree.
T he convergence of the OSPF protocol is extremely fast, compared to all other internal gateway protocols. It is consisted of three factors, which must be careful taken into account, when designing an OSPF network. These are (Goyal et al., 2012):
Detection of a topology modification – this is the time needed by OSPF to detect a link or interface change or even worse failure.
Establishment of a new adjacency or revocation of an existing one – in response to a change in the network.
Propagation of a change in the network – this time entails the generation of LSA messages and their flooding throughout the network (or better the area).
SPF tree calculations – this is the time spent by each router in order to run the SPF algorithm and provide a loop-free topology.
Forwarding table creation – this is the time needed for the router to create the routing table.
Thus the total time needed for OSPF to converge is: Convergence Time = Change Propagation Time + SPF Execution Time + Routing Table Creation Time + Failure Detection Time In a typical convergence situation the average time needed for a router to propagate the Link State Advertisements and run the SPF algorithm is slightly less than 1 second. In parallel, the default time for the SPF algorithm to rerun (delay timer) is 5 seconds. This provides the lower boundary for the convergence of the OSPF protocol in its default settings. The upper boundary is determined by factors such as the size of the networks, size of the topology database and of course, the type of failure. In the worst case, a link fails without an alternative route existing, thus leading to the protocol waiting for the dead timer to expire introducing a 40 second delay in the default situation.
D epending on the network type, mainly in multi-access networks, the OSPF process may lead to the election of a DR (Designated Router) and a BDR (Backup Designated Router). These roles are the focal points for the exchange of OSPF information, reducing the full adjacency relationships between routers connected on a multi-access medium. Each non-DR or non-BDR router forms a full adjacency relationship only with the DR and BDR, exchanging routing information only with these two routers on the network segment. Therefore the role of the DR is to distribute the updated topology information to all routers on the same segment leading to a significant reduction in routing traffic. In order to accommodate the aforementioned process, two multicast IP addresses are used: 18.104.22.168 is used by all routers on a segment to inform the DR and BDR on any topology changes, while 22.214.171.124 is used by the DR to send Link State Updates to all routers on the segment.The factors that determine the router that wins the DR/BDR election process are the following:
1. The router with the highest priority becomes the DR on the multi-access segment. By default all routers have a priority of 1 and a priority of 0 enforces the router not to participate in the election process. 2. The router with the highest Router ID becomes the DR. In this point it must be mentioned that the Router ID is determined in order of significance by: i. the commend router-id which sets the Router ID to a particular value or ii. the largest IP address of the loopback interfaces configured on the router or iii. the largest IP address of the active interfaces configured on the router. The router with the second highest priority or Router ID becomes the BDR.
In order to enforce the stability of the OSPF process: When a BDR becomes DR, a new election takes place so as a BDR to be elected. The factors presented above are taken into account when electing a new BDR. If a router is being introduced in the network with highest priority after the election of the DR and BDR it cannot be elected until one of the DR and BDR routers fail (Cisco, 2005).
O SPF is a hierarchical routing protocol segmenting the entire network into smaller areas. These are logical groups of routers, thus providing the capability of reduced instability, topology update containment as well as shorter routing tables. It must be stressed that when a network is constituted by more than one area, the OSPF protocol imposes some restrictions. Specifically, OSPF is composed of a centralized backbone network the Area 0 which links all the other lower areas within the hierarchy. All these areas must be physically connected to the backbone area in order to exchange routing information (each area stores a distinct link-state database). This is forwarded to the backbone area which afterwards floods this information to the other areas, thus decreasing the traffic between the different parts of the autonomous system. It must be noted that an area can be implemented either in the IP address format (0.0.0.0) or the decimal format (0). In the case that an area is not physically connected to the backbone area, a virtual link is required to be configured (Graziani & Jonson, 2008; Ferguson & Moy, 2008). In order to accommodate better configuration and tuning of the protocol, different area types have been defined. These are: i.Backbone Area The Backbone Area constitutes the logical and physical core structure of an OSPF network and is accountable on distributing all routing information between non- backbone areas while it is situated at the center of all areas. It must be noticed that backbone must be contiguous and the connectivity could be established and maintained through the configuration of virtual links. Moreover, all OSPF areas have to be directly connected to the backbone area even through a virtual link. ii. Stub Area Stub Area is restricted in receiving route advertisements external to the autonomous system (AS), therefore the database size is reduced even more. Even so, stub area receives information about networks from other areas of the same OSPF domain. Basic features that are associated with stub area are elaborated below: Stub area permits inter & intra area routes. Stub area prohibits flooding of external LSAs. A default route is defined inside stub areas. OSPF routes inside the stub area must configured as stub routers. iii.Not-So-Stubby Area (NSSA) A Not-So-Stubby Area (NSSA) is a continuation of stub area that allows autonomous system external route infiltration into the stub area sending them to other areas without being capable of receiving external routes from other areas as well as importing external addresses. Distinguishing features of Not-So-Stubby Area are quoted below: Autonomous System Boundary Router injects Type 7 LSA external addresses. Area Border Router Type7 LSAs are converted into Type5 LSAs at NSSA, which are flooded to other areas. NSSA allows summary LSAs. NSSA prohibits external LSAs. iv. Totally Stubby Area A Totally Stubby Area is resemble to a stub area being physically connected to the backbone area from which only receives default route. Totally stubby area reaches other networks by a default route which is the only Type-3 LSA advertised into the area. Certain characteristics are presented as following: Summarized routes such as Inter-area (IA) are not allowed into totally stubby area. Intra-area routes are prohibited into totally stubby area. The default route is permitted as a summary route reducing system resource usage due to the fact that the route processor supports less routing decisions. v.Totally (NSSA) Not-So-Stubby Area A Totally Not-So-Stubby Area is a combination of a TSA and a NSSA where only the default route is allowed from the backbone area (0.0.0.0) as well as the injection of external information in the local area with the ASBR and traverse. An area could be characterized as totally and NSSA presenting the features as following: Summarized routes Type 3, 4 or 5 LSAs are not flooded in totally NSSA. External routes are prohibited except the default route as summary route. Area Border Router Type7 LSAs are converted into Type5 LSAs at NSSA, which are flooded to other areas (Shamim, et al., 2002). vi. Transit Area The transit area includes two or more OSPF border routers which get through network traffic from one adjacent area to another.
In the following table the main advantages and drawbacks of the OSPF protocol have been summarized.
T he Enhanced Interior Gateway Routing Protocol is a CISCO dynamic proprietary protocol for the Internet Protocol (IP), IPX and Appletalk networks designed by CISCO Systems at the University of California at Santa Cruz in 1992. However, in 2013 CISCO published it as an open standard. EIGRP belongs to the distance vector routing protocol family characterized as the more advanced of its kind due to the fact that it is more scalable in medium and large scaled networks. Despite the fact that it belongs to the distance vector family, it carries link state protocol features and is publicly characterized as a hybrid distance vector protocol. It must be stressed that is used in order to distribute routing information within the same Autonomous System sending incremental updates, minimizing the amount of work on the router as well as of data that is required to be transmitted. The most important feature of EIGRP that should be stressed is that it uses both equal-cost load balancing (ECLB) as well as unequal-cost load balancing. The former takes place in the same way as in the IGRP and OSPF. In other words, in networks with multiple equal-cost paths to the same route destination, rationally load-share traffic equally among these paths occurs. However, EIGRP is the only protocol that makes intrinsically equal & unequal cost load balancing. This happens through the use of the variance parameter. The EIGRP has the ability to combine successor routes with feasible successor routes (that exist in the topology table) in order to implement unequal-cost load balancing (Cisco, 2005; Albrightson, Garcia-Luna-Aceves, & Boyle, 2011).
The following table presents the encapsulation of an EIGRP packet inside a Data Link frame (Cisco, 2007).
More specifically, the EIGRP Packet Header is encapsulated in an IP packet with a Protocol field of 88 while the destination address is the multicast address 126.96.36.199. The EIFRP Packet Header is depicted in the table below (Leahy, 2015):
Furthermore, EIGRP uses six different packet types. Each one serves a different purpose. These are (Leahy, 2015): Hello: Are sent via multicast in order to identify neighbors (unreliable) Acknowledgment: Are sent via unicast to confirm reliable delivery of EIGRP packets (unreliable) Updates: Are sent via RTP & unicast in order to convey reachability of destinations (reliable) Queries: Are sent via RTP & multicast requesting routing information, for instance the status of the route for fast convergence (reliable) Replies: Are sent via RTP & unicast in response to Query packets (reliable) Requests: Are sent via multicast or unicast in order to collect distinguish information about the neighbors (unreliable)
EIGRP associates six different vector metrics with each route and takes into account only four of them in order to compute the composite metric. Bandwidth: Minimum Bandwidth along the path from a router to the destination. Load Number: which ranges from 1 to 255 Total Delay: Delay along the path from a router to the destination. Reliability Number: which ranges from 1 to 255 MTU: Maximum Transmission Unit is never used in the metric calculation. Hop Count: Number of routers a packet passes through the network. Ho count is never used in the metric calculation.
E IGRP computes routing metrics using the minimum bandwidth on the path to a destination network as well as the total delay. It must be stressed that four vector metrics such as bandwidth, reliability, delay and load are being associated in order to compute the Composite metric for the determination of the preferred route (successor). The minimum bandwidth and the total delay metrics are defined from the values that have been set up on the interfaces of the routers in the path to the destination network using the following formula in order the EIGRP routing metric to be calculated (Cisco, 2015; Albrightson, Garcia-Luna-Aceves, & Boyle, 2011):
The default values for K weights are: K1 = 1, K2 = 0, K3 =1, K4 = 0, K5 = 0, considering that the K2, K4 and K5 weights are zero by default, effectively the EIGRP metric formula leads to proceed in the following form: (bandwidth + delay) * 256Where Bandwidth and Delay are valued with the following calculations:Bandwidth = 107 / Value of the bandwidth interface command (Constitutes the link with the least amount of Bandwidth).Delay = Value of the delay interface command (Is related to each interface in milliseconds and becomes cumulative while a specific route crosses router after router).
E IGRP supports IPv4 classless addressing and utilizes the DUAL algorithm in order to create the routing table. Both the algorithm and data structure (Neighbor Table & Topology Table) will be analyzed below:
EIGRP routers obtain information about the state of the adjacent neighbors and their IP addresses. Every time new neighbors are discovered their IP address and interface are recorded and stored in the neighbors’ table (data structure). While the neighbor sends Hello packets, it also advertises the Hold Time to determine whether the neighbor is operational and reachable. It must be noted that the ASN (Autonomous System Number), Subnet Number and K values must be identical in order for the neighbor adjacency to be formed. Hello packets are sent to the multicast address every 5 seconds on LAN interfaces & every 60 seconds on WAN interfaces to verify that the neighbor relationship is still active. If the Hold Time Interval passes (hold-down timer by default is 15 seconds) due to the fact that a Hello packet wasn’t heard within this, the DUAL algorithm is forced to run taking into account the topology changes. Furthermore, the neighbor table contains essential information for the RTP (Reliable Transport Protocol) mechanism in order to pair acknowledgements with their corresponding data packets. It must be stressed that round trip timers are stored in the neighbor table in order to evaluate an optimal retransmission interval (Graziani & Jonson, 2008).
EIGRP uses the DUAL (Diffusing Update ALgorithm) or else DUAL FSM (finish-state machine) which ensures that each route will be loop-free calculated in order for routing loops to be avoided. This algorithm responds promptly in changes that might occur in the routing topology and adjusts dynamically the routing tables. The factors that contribute in the loop-free routes mechanism are being analyzed below (Xu, Dai & Garcia-Luna-Aceves, 1997):
i. Feasible Distance (FD) is the best EIGRP metric or else the lowest cost along a path to a destination network with the participation of the route metric that has been advertised by the neighbor, listed in the routing table. ii. Reported Distance (RD) / Advertised Distance (AD) is the total cost of the route as advertised by the neighbor & needed along the path to the destination network. iii. Successor also known as current Successor (or primary route) is the route with the lowest Feasible Distance guarantying a loop-free path to a destination. The successor routes are installed in the routing table in order to be used for forwarding packets. iv. Feasible Successor (FS) is the backup route with reported distance less than the feasible distance. The FD of the Feasible Successor is greater than the FD of the Successor, however it’s Advertised Distance (AD) must be lower than the FD of the Successor. These routes are stored in the topology table and are promoted immediately when the Successor route fails. v. Feasibility Condition (FC) is the condition that provides loop-free routes to a destination with the contribution of the Successor and Feasible Successor routes. Feasibility Condition states that the Reported Distance must be lesser than the Feasible Distance [RD < FD] in order for a route to become a feasible Successor (Cisco, 2015).
EIGRP topology table includes all learned routes to a destination advertised by neighboring routers. Specifically, the topology table stores routes and their metrics, Successors and Feasible Successors as well as locally connected subnets. It must be noticed that routes in the topology table are usable by the router only when they are active and inserted into the routing table or have a higher AD than an equivalent path. For every reachable network, the topology table contains the total delay, reliability and path loading, the lowest bandwidth on the path (the weakest link), the feasible and reporting distance and finally the route source (Graziani & Jonson, 2008).
C onvergence starts when two routers become neighbors. This dynamic learning happens through the exchange of hello packets (default hello timer is 5 seconds on high-bandwidth links and 60 seconds on slower links). The outcome of this neighbor discovery is the creation of the neighboring tables with all the additional features as described in previous sections. At that point the neighboring routers exchange routing information and build their corresponding topology tables. In a next step they employ the DUAL algorithm to calculate the feasible and reported distances, and of course the successor and feasible successor routers. The latter routes may exist in the case the feasibility condition is met, thus providing loop-free alternatives to the successor route. The feasible successor routes and their existence is utmost significant to the EIGRP convergence process. When a successor (primary route) fails, then the EIGRP process (Sankar & Lancaster, 2014): checks for a feasible successor and if one is found then it is immediately promoted to a successor and is inserted in the routing table If a feasible successor does not exist, then the EIGRP process marks the failed route as ACTIVE in the topology table and starts sending query packets to all neighbor routers to find an alternate route to the network that failed. If these neighbors do not have an alternate route then they mark this failed route as ACTIVE in their topology tables, and generate query packets which they forward to their neighboring routers and so on. In case a router knows an alternate path, he responds to the query packets and all routers converge through a recursive process. In the case no router responds, the routers keep this route as ACTIVE until the corresponding EIGRP timers expire, but until then they all are Stuck-In-Active (SIA). The aforementioned convergence process poses a threat to the scaling of an EIGRP network in an arbitrary way. When the number of routers in an EIGRP network grows to the number of hundreds, then the stuck in active situation may bring the network to its knees. In that case, a strict design must be implemented both in organizational structure and in route summarization. Summarizing all the above points, in order for a network designer to make EIGRP convergence quicker he must: Implement shorter timers. In this way, routers can form relationships faster and detect dead neighbors more efficiently. Provide Route Summarization through the hierarchical structure of the network. In this way, when a query arrives at an EIGRP router which features a summarized route, he immediately replies to this query, thus terminating quicker the stuck-in-active situation. Configure route filtering, so that a router will immediately respond to an EIGRP query with an inaccessible message reply, terminating again the SIA and helping in removing a non-existing route from all routing tables. Configure stub routers in remote locations so that central routers not require forward any queries to these.
In the following table the main advantages and drawbacks of the EIGRP protocol have been summarized.
E IGRP and OSPF dynamic routing protocols share various fundamental characteristics. They both implement Variable Length Subnet Mask (VLSM) as well as Classless Inter-Domain Routing (CIDR) not to mention they are intrinsically designed to accomplish fast convergence and backup routes in link failure situations. Furthermore, OSPF and EIGRP are capable to handle their own routing tables and send partial updates when changes occur in order to achieve lower traffic on the network. Despite the fact that these two popular protocols have such possibilities a summary criteria based comparative study at least from theoretical approach will highlight briefly the advantages and disadvantages of each protocol.
EIGRP implements a combination of metrics for the routing estimation to the destination network which is based mainly on bandwidth and delay and optionally on load, reliability & MTU in comparison to OSPF which calculates a route to a destination network based on the cost. This shows that EIGRP has an advantage of the wider management of the network traffic over OSPF.
If changes occur in the network topology each protocol in order to achieve fast convergence, must proceed to the recalculation of the route to the destination network. From literature approach Ayub, et al. (2011) and Islam & Ashique (2010) demonstrate through simulation experiments that EIGRP has faster convergence over OSPF as the second researchers through the obtained results argue that EIGRP convergence is about 6 seconds faster than the OSPF. Though, in conditions of administrator interference in timer values OSPF seems more efficient.
It is ambiguous which protocol provides higher network throughput. It is known that EIGRP as the hybrid protocol could operate both as distance vector and link state depending on the demands of the network topology and the DUAL algorithm. Based on this peculiarity of the protocol Thorenoor (2010) states that the EIGRP makes visible the better CPU usage as well as the bandwidth control. Yet, after throughout simulation testings between EIGRP and OSPF, Islam & Ashique (2010) claim that OSPF accomplishes better network throughput. Due to the fact that network topologies in each simulation test may change and thus the results, it makes it impossible to say with certainty which protocol overrides the other.
The EIGRP routing protocol is planned to operate on flat network topologies while the OSPF hierarchical protocol in these network topologies increases CPU usage and memory requirements due to the fact that it has to contain every node in the routing table. However, OSPF is considered more robust protocol and with attentive configuration could reduce the size of the routing table and respond with consistency to a scalable network topology. Through simulation experiments the stability of all those reported so far for the OSPF and EIGRP dynamic routing protocols are expected to be demonstrated.
S imulation is the method that is followed for testing or research purposes representing a real system through virtual reality. Modeling through a simulation tool is the beneficial way to conduct experiments in virtual environments which otherwise might be impractical because of the equipment required, the high cost needed to be spent or even due to the fact that the system might not support extensive tests. A simulation tool is a computer-based mathematical software which performs multiple algorithms and equations in order to render results based on the input data. This permits the examination of complex behavior as well as scenarios on a wide range of conditions fast and easy than in a physical environment. In the case of this thesis, the output results will be obtained using a Computer-Based software Simulation which is called the Riverbed Modeler Academic Edition 17.5. Suitable equipment will be applied as well as necessary procedures in order to measure the performance of the OSPF and EIGRP routing protocols based on the desired quantitative metrics (Prokkola, 2008).
R Riverbed Modeler Academic Edition 17.5 is considered a high-speed software providing performance management for networks and applications. This software previously known as OPNET, is a simulator that constructed based on the Discrete Event System (DES). It simulates the system behavior by modeling the events in the system defined by the user’s processes according to the Riverbed Splash Community. It provides model designs, data collection and data analysis not to mention multiple simulation iterations. Unfortunately, the academic edition of the software mostly supports teaching purposes without providing extensive research area which alters the depth as well as the expandability of this thesis. The primary plan was to examine and compare the behavior between EIGRP and OSPF in a scaling network, consisting of multiple subnets adapted in every possible topology, to generate full and partial mesh networks with more than a hundred nodes as well as to implement the entire potential scenarios with link aggregation and redundancy which was impossible. In order the output results to be accurate, the network topologies based on the software’s restrictions will include less nodes than the primary plan and will be linked in series resembling the network diameter.
The hierarchical Structure of the Riverbed Modeler is classified in three main domains. i.Network domain The network model depicts the overall system which is composed of the network and the possible sub-networks including also network topologies, physical connections, interconnections, geographical coordinates and configuration.
ii.Nodes domain The nodes domain constitutes the interval substructure which could be a workstation, a group of routers, mobile devices or even a subnet consisted of servers, switches, client computers, hubs, remote sensors, satellite terminals and more.
iii. Process domain In this field the type as well as the load of traffic are determined in order the equivalent results to be generated. The process domain constitutes the source code as well as the single modules inside the network nodes such as the IP protocol or the data traffic source model.A very significant feature of this simulator that should be noted is that fact that it provides the possibility of running external code components through the External System Domain (ESD) (Prokkola, 2008).
In order to achieve the simulation-based comparison between the OSPF and EIGRP routing protocols specific stages in process have to be followed for the most accurate design of the simulation tool. The sequence of steps is depicted below:
T he dynamic routing protocols that will be examined in this thesis are the OSPF and EIGRP in IPv4. It must be noted that they will be compared and evaluated in terms of performance based on the quantitative metrics which are the network convergence duration, video conference packet delay variation, IP voice jitter and CPU utilization. This simulation will present which one of the two protocols prevails over the other based on their performance by applying real traffic in the entire network.
In order to evaluate the performance of EIGRP and OSPF dynamic routing protocols, four scenarios were created and implemented in two different network topologies. The first one is the Base Topology which consists of seven routers and the second one is the Scaled Topology which is composed of ten routers. Through these two different topologies, this thesis will indicate the scale effects which may arise. Both network topologies were originated on a Campus layout (20X20) with multimedia users on the one side as well as voice and multimedia servers on the other side. It should be noted that the total time duration to run the simulation was set to five minutes. This occurs due to the fact that the Riverbed Modeler Academic Edition 17.5 presents many intrinsic limitations on the number of the events and creating a more complex topology would depict non accurate results. One other significant thing regarding OPNET that should be mentioned is that every scenario runs separately so as to yield accurate average values.
The Base Topology constitutes the base line of the initial results between OSPF and EIGRP performance. As shown in Figure 7 this topology has two serial routes where the top line is PPP DS3 link with fewer routers and the bottom line is PPP DS1 link with one additional router. It is clear that the traffic will pass over the top route. Hence, in order to draw the results of EIGRP and OSPF performance, a link failure will occur between R2 and R3 so as to observe how each protocol behaves.
The Scaled Topology follows the same philosophy with the Base network topology, however is considered more complex. Specifically as shown in Figure 8 it includes two main routes where the top line is PPP DS3 link with fewer routers, the bottom line is PPP DS1 link with one additional router and the middle lines are also PPP DS1 link. The only difference is that it has increased the number of the routers and installed a router in the middle which is connected to the R2, R3, R5 and R6 routers. As in the Base topology a link failure will occur between the R2 and R3 routers which will reflect the performance results of EIGRP and OSPF protocols in a more complicated topology. In both of these topologies the link failure between R2 and R3 begins in the second minute (120 seconds) and remains down causing disturbance in the network routing until the third minute (180 seconds), where link recovery occurs and connection returns to the initial state. Equally Base and Scaled topologies are composed of the network devices and the configuration utilities as follows: IP Routers Ethernet Servers Ethernet Switches PPP DS1 Links PPP DS3 Links Ethernet 10BaseT Links Ethernet 100BaseT Switch LAN Application Configuration Profile Configuration Failure Recovery Configuration Furthermore, both Base and Scaled topologies include in their workspace the Profile Definition Object which is named Users and the Application Definition Object which is called Apps Traffic. Profiles portray the activity patterns of a client or group of clients in terms of the applications utilized over a range of time. In the Profile Object Definition of this thesis multimedia clients are created in order to support Video Conference and Voice over IP (VoIP). It must be noted that Profile Definition Objects are related to Application Definition Objects due to the fact that users’ profiles are constructed using different application, where several types of traffic as well as usage parameters could be specified. Concerning the requirements of this thesis, the Application Definition Object is adjusted to support Video Conference and Voice over IP.
Both EIGRP and OSPF dynamic routing protocols will be applied in all routers of the two different network topologies with the exact same infrastructure for each protocol. Specific quantitative metrics were chosen in order to measure the performance as well as to evaluate the behavior of these protocols in every scenario. First, the network convergence duration will be measured, then the packet delay variation for video conference, after that the jitter of the voice and finally the CPU Utilization. The last scenario will demonstrate in what extend the first router (R1) located next to the multimedia clients faces difficulties during the traffic that passes over it.
From Figure 9 it is observed that the average convergence duration of OSPF is faster than that of EIGRP on the Base Topology. This means that when an alteration emerges in the OSPF network, the routing table is recalculated and all routers within the area update the topology database by flooding LSAs to the neighbors while in the EIGRP network the routers send queries to the direct neighbors in order to propagate the updated routing table where the successor has been recalculated.
When the network grows, it is noticed that the OSPF and EIGRP have about the same performance both in Base and Scaled Topology regarding the network convergence duration. Particularly, the average convergence duration of OSPF is presented faster in Figure 10 where unlike the Base Topology the EIGRP protocol seems even slower.
The average delay variation is estimated between the beginning of the packet transmission from the source until its reception to the destination. Figure 11 indicates that OSPF reflects greater average packet delay for video motion thus has a lower throughput in comparison to EIGRP which seems to have better performance concerning the average packet delay variation on the Base Topology.
The results of the Scaled Topology appear in Figure 12 approximately the same as those of the Base Topology where the EIGRP protocol has less average packet delay variation hence larger throughput compared to OSPF.
IP Voice jitter is determined exclusively as a variation in the delay of received voice data packets affecting the voice quality as well as the data. Figure 13 imprints that OSPF network in Base Topology presents utmost higher average IP voice jitter compared to EIGRP.
The particular Figure 14 shows that the OSPF network in the Scaled Topology has slightly average higher jitter from EIGRP.
In terms of average CPU Utilization, the simulation results in Figure 15 depict that OSPF network requires more processing power compared to EIGRP network.
In comparison to the Base Topology, Scaled Topology shows via Figure 16 that when the network scales and more routers are added, EIGRP demands more computational resources compared to OSPF network.
The section of this thesis will present the results of the four scenarios implemented together in Base and Scaled Topology being critically evaluated.
This Figure 17 clearly shows that the average network convergence duration of EIGRP in scaled networks is much lower than OSPF for real traffic, which is opposed to more researches and simulation processes. How fast the convergence is achieved depends on the number and the size of the updates that are being sent. Probably this effect is due to the specific nature of the Base and Scaled Topology where EIGRP in link failure situations is not as effective as OSPF with the absence of feasible successor.
As it seems in Figure 18, OSPF in both Base and Scaled Topology presents higher average packet delay variation, thus lower throughput compared to EIGRP, where data packets reach faster to the destination. This possibly occurs due to LSAs flooding which consume more bandwidth. It must be noted that both of these protocols use equal cost paths (the same bandwidth on the links).
Through the Figure 19 is shown that the OSPF has higher jitter on both topologies compared to EIGRP which of course as depicted, launches in Base Topology. Probably when the link failure occurs in the second minute and OSPF recalculates the optimum path in order to reroute the traffic, the network is flooded with LSAs with impact the high IP voice jitter.
Figure 20 indicated that OSPF in the Base Topology imposes a CPU and memory burden on the router (R1) in comparison to EIGRP which demands less bandwidth and CPU memory utilization probably due to the fact that initially OSPF runs more processes than EIGRP in order to route the packet. Of course when the network scales, it is observed that EIGRP requires slightly more resources compared to OSPF due to the fact that EIGRP faces difficulties in random network failures which pushes it to work more intensively. Perhaps because of the topology requirements (relative position of routers and cost), EIGRP had to send more queries and replies.
T he OSPF and EIGRP interior dynamic routing protocols are to the greatest degree widely implemented in most networking infrastructure. Through this thesis, a simulation-based comparative study was conducted in order to indicate which of the above mentioned protocols dominates according to specific quantitative metrics. After the extensive literature review, the exhibition of both protocols features and the simulation implementation, this thesis critically evaluated all the information and collected the simulation results in order to point out which protocol has optimal performance.Most researchers insist that the EIGRP protocol has better performance especially on network convergence duration and CPU memory utilization compared to the OSPF protocol. The hypothesis of this thesis attempted to present that this does not appear to be true if the large network is not designed with a hierarchical structure which was confirmed partially.
Base and Scaled Topology were created where both EIGRP (Distance-Vector) and OSPF (Link-State) protocols for IPv4 were applied in order to present the impacts on simulated networks, the behavior of the protocols as well as the scaling effects. In the opposite direction of other researchers who claim that EIGRP has faster network convergence duration compared to OSPF, the simulation results indicated that in both Base and Scaled topology OSPF converges more quickly compared to the EIGRP routing protocol in these specific topologies. In terms of packet delay variation, the performance of EIGRP and OSPF was measures based on real time traffic via video conference. It was observed that the packet delay variation of OSPF is higher opposed to EIGRP, due to the LSAs flooding thus more bandwidth consumption occurs. This of course impacts the throughput of OSPF which is lower compared to EIGRP.
Concerning IP voice jitter, the OSPF protocol presents higher voice packet delay compared to EIGRP, due to the LSAs flooding. Regarding the last scenario, results were different in both topologies. In other words, in the Base topology it was observed that the OSPF required more computational resources as opposed to Scaled topology where EIGRP demanded slightly more resources compared to OSPF. This is due to the fact that the EIGRP needed to send more queries & replies. It should be noted that this was the only scenario that showed different results of the EIGRP and OSPF protocols in scale effects. Although the aim of this thesis was achieved, limitations of the Riverbed Modeler Academic Edition 17.5 simulation tool did not permit further research and analysis of larger and more complex topologies. Because of these restrictions it is difficult to judge which of the two protocols is best in terms of performance where many factors acquiesce. It must be stressed that many factors play a crucial role upon the selection of the protocol that will be used in each circumstances such as the infrastructure, the network size and the requirements to be satisfied every time.
Part of future work could be extended in studying the implementation of several types and more demanding traffic as well as with heavy load and multiple sudden network failures. Furthermore, experiments with large network scaling could be conducted in order to highlight the multi-area in the OSPF routing protocol. Finally, researchers could proceed in extensive experiments of EIGRP and OSPF in IPv6 using a professional Research and Development (R&D) application.
D uring the conduction of this thesis, I gained many things. First and most significant, I studied in depth both EIGRP and OSPF dynamic routing protocols learning features and functions that up before I ignored their existence. Moreover, I learned how to perform an integrated technological research not to mention how to structure a thesis. It should be noticed that I faced troubles with the simulator due to the fact that it was an Academic Edition with various restrictions confining me considerably from providing a more thorough dissertation thesis. Finally, it must be stressed that this experience was time consuming but also efficient because it helped me to learn how to manage my available time correctly which will follow me in my entire life.