The strengths and weaknesses of redistributing between two different interior routing protocols

Introduction

Complex enterprise networks are often made of multiple routers running one or more dynamic routing protocols for a routing information exchange as well as typically containing some static and directly connected routes. A router running an IGP (Interior Gateway Protocol) such as OSPF, EIGRP, RIP or ISIS can only exchange the routing information with its peers running the same kind of IGP under the same instance (therefore the route redistribution from OSPF into a different OSPF instance is feasible). Due to various reasons e.g. merge or split of two independent routing domains the need to exchange a routing information between two different routing protocols may arise. This problem can be resolved by implementing the route redistribution. In this document, we look at the route redistribution between the Open Shortest Path First (OSPF) and the Enhanced Interior Gateway Routing Protocol (EIGRP) routing protocols and their associated strengths and weaknesses. These two diverse protocols selected due to their dominance in today’s computer networks and the fact that they belong to a different major protocol class – OSPF representing the Link-state and EIGRP representing the Distance-vector protocol class.

Strengths and weaknesses of redistributing between OSPF and EIGRP protocols

While route redistribution provides great benefits by allowing to connect two separate routing instances and exchanging the routing information in a dynamic fashion, which otherwise would not be possible it can also introduce some unwanted behaviour especially when designed and implemented carelessly.

Among the strengths are the abilities of both protocols to manipulate parameters and influence the routing decisions. However, there are some important weaknesses to consider such as different convergence times, as Thorenoor, S. (2010) and Wijaya, C. (2011) points out EIGRP convergence times are much faster than OSPF which can cause convergence issues.

Due to OSPF and EIGRP being so different protocols and using different algorithms and metrics in their best path determination process it is not possible to dynamically exchange the route metric information during the redistribution process. Should the routes be redistributed without manually specifying the metric they would get the default seed metric assigned, when redistributing into OSPF the routes are assigned a type 2 (E2) metric of 20 (with an exception when redistributing from OSPF into OSPF and the original metrics are preserved), and when redistributing into EIGRP the routes are assigned a metric of 0 which is interpreted as unreachable by the protocol and tells the router that the route should not be advertised. In the OSPF example the E2 route metric is fixed and will never change when further propagated, therefore if such route is redistributed from EIGRP into OSPF domain and the local routes have a higher metric it may result in a sub-optimal routing. The metric, the metric type and an Administrative Distance (AD) has to be taken into consideration and manipulated when redistributing the routes.

As an example, when the desired route is learned by redistributing from OSPF into EIGRP, the route is marked as D EX – EIGRP external route and has the AD of 170, therefore the internal route is preferred due to its lower OSPF AD value of 110, if required this behaviour can be adjusted by using a route map to change the distance when the route is redistributed.

In a large networks and complex topologies with redundant entry and exit points between the two routing domains and especially when a multipoint mutual redistribution takes place the risk of self-sustained routing loops and suboptimal routing is very high. Even through if the routing loops are not initially forming after the redistribution is configured, an extensive analysis carried out by Vissicchio et al., (2014) shows that the configuration changes or activation of route redistribution alone can cause prolonged routing loops and network instability across the domain.

“Even in a small network like Geant (35 routers), up to 31 loops occur during the reconfiguration exclusively because of the activation of route redistribution, with a median around 20 loops per experiment. In addition of being numerous, many loops are also long-lasting. In median, a loop lasts for about 15% of the reconfiguration process, while some loops span about 50% of the reconfiguration. As expected, the number and the duration of loops grow with the size of the network. In Rocketfuel topologies, between 100 and more than 500 loops are raised in the worst case. Even worse, loops also last longer in Rocketfuel topologies, with peaks reaching more than 80% of the reconfiguration process. Depending on the network size, those results imply that reconfiguration loops can last in real networks from minutes to hours, even if the reconfiguration of a single router only takes few seconds.”

As pointed out by Vissicchio et al., (2014) and demonstrated in Fig. 1. There is a positive correlation between the network size and the number of routing loop occurrences and their duration during and after a network configuration change.

The following tools, techniques and a good understanding of the network topology, the related routing protocols’ operation and their convergence times can mitigate such unwanted behaviour:

Preferring other redistribution techniques than multipoint two-way redistribution.
Redistributing only internal routes.
Advertising only the default route.
Tagging incoming routes and not allowing them to be re-redistributed back into originating routing domain by filtering outgoing routes based on the tag at the redistribution exit point.
Using tools such as route-maps, prefix-lists, ACLs and distribute-lists to implement the routing policy.
Not using the default seed metrics.
Increasing the metric of the redistributed routes to be the higher metric than any internal routes.
Manipulating Administrative Distance.
Using distribute-lists combined with an ACLs to stop advertising unwanted routes.
Using ‘subnets’ and ‘include-connected’ keywords with the ‘redistribute’ command to advertise the classless and connected subnets.
Conclusions

The route redistribution can be implemented in various ways depending on a network design and the desired outcomes, it may also require a careful planning and implementation as well as a sound understanding of the routing protocol operation and the network topology to ensure that a sub-optimal routing and a self-sustained routing loops are not introduced into the network and any unwanted behaviours are mitigated.

Reference List

Stefano Vissicchio, Laurent Vanbever, Luca Cittadini, Geoffrey G. Xie, Olivier Bonaventure (2014) ‘Safe Routing Reconfigurations with Route Redistribution’, IEEE INFOCOM 2014 - IEEE Conference on Computer Communications. [Online], Available at http://ieeexplore.ieee.org.libezproxy.open.ac.uk/stamp/stamp.jsp?arnumber=6847940 (Accessed 10 September 2016).

Alim, A. (2011) On the interaction of multiple routing algorithms [Online], CoNEXT '11: Proceedings of the Seventh COnference on emerging Networking EXperiments and Technologies, ACM, Available at http://dl.acm.org.libezproxy.open.ac.uk/citation.cfm?id=2079303&CFID=666362052&CFTOKEN=57742071 (Accessed 11 September 2016).

Thorenoor, S. (2010) Dynamic Routing Protocol implementation decision between EIGRP, OSPF and RIP based on Technical Background Using OPNET Modeler [Online], IEEE. Available at http://ieeexplore.ieee.org.libezproxy.open.ac.uk/stamp/stamp.jsp?tp=&arnumber=5474509 (Accessed 14 August 2016)

Wijaya, C. (2011) Performance Analysis of Dynamic Routing Protocol EIGRP and OSPF in IPv4 and IPv6 Network [Online], IEEE. Available at http://ieeexplore.ieee.org.libezproxy.open.ac.uk/stamp/stamp.jsp?tp=&arnumber=6141697 (Accessed 14 August 2016)