Lesson 34 - Dynamic Routing Protocols Introduction

If you have read the previous post, you must have noticed that using a static routing method tends to be a bit cumbersome in larger implementations. Using one of the dynamic routing protocols feels like an easier solution in these scenarios.

In this post I will briefly explain the general concepts behind dynamic routing protocols. Then, we can jump to implementation fundamentals.

One way of classifying dynamic routing protocols is based on where they are used. This criterion allows us to distinguish between two major solutions:

1. Interior Gateway Protocols  (IGP) 
2. Exterior Gateway Protocols (EGP)

Common Interior Gateway Protocols are: 

• Routing Information Protocol (RIP), 
• Open Shortest Path First (OSPF), 
• Enhanced Interior Gateway Protocol (EIGRP, Cisco proprietary protocol), 
• Intermediate System to Intermediate System (IS-IS).

Exterior Gateway Protocols (currently there is only one in use)

• Border Gateway Protocol  (BGP)

IGPs are designed to work in private networks. EGPs are used to provide paths in the public network (Internet). 

We can also classify routing protocols based on the algorithm they use to distribute and maintain information (routing table). There are three major algorithms supported by Cisco routers:

1. Distance Vector (DV, aka Bellman-Ford) – example of protocol: RIP.
2. Link-State – example of protocols: OSPF, IS-IS.
3. Advanced Distance Vector – protocols: EIGRP (also BGP is partly distance vector protocol).

Understanding the algorithms helps us determine the proper solution for a given design. There is no one best routing protocol out there, but there could be the best one in a specific design.

In this post I am going to focus in on the first algorithm listed above.

Distance Vector Algorithm Characteristics

This method is sometimes referred to as ‘routing by rumor’. The main characteristics of this approach are:

• Routers do not know the topology of the network. They only know which is the outbound interface and the next-hop router’s IP address (vector) as well as the metric value which describes how far the destination is (distance).
• Routers advertise their full routing table periodically. This method of route distribution creates two problems: routing loops and counting to infinity. Special techniques were created to solve these issues (details later in the post).
• Routers perform automatic summarization if they are connected to different classful (A, B, C) networks.
• No VLSM support. All network masks must be identical if the subnets of a major class are used in the network (RIPv1). RIPv2 is classless (VLSM supported using ‘no auto-summary’ command).
• Routers are slow to converge. It takes a lot of time to invalidate lost routes and pick the new path if one is available as well as to synchronize their routing information.
• Routers use simple metric. The metric number tells a router how many routers the packet has to traverse in order to reach the destination. In modern networks bandwidth of the path is much more important than how many hops will be used.

The above characteristics do not encourage us to use this kind of solution in our modern networks. But knowing the DV rules help us appreciate protocols such as OSPF or EIGRP which are more likely to be used in our designs.

Let’s see how things work when DV algorithm is used. As an example, I will use RIP protocol hoping to explain the principles of operation and how the two design issues have been solved (routing loop and counting to infinity).

Distance Vector Principles of Operation

Consider this simple topology. Without getting into configuration (syntax) details let’s have a quick discussion on how information is distributed using DV algorithm. Initially, the routers recognize only connected subnets. They are populated in the routing table as soon as IP addresses and network masks are configured and they are activated (no shutdown).

Pic. 1 - Connected Subnets.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Let’s assume that we have enabled RIPv2 protocol in the topology presented above (pic. 1). This version of RIP allows the routers to announce both the subnet IP addresses and the network masks (we’ll put it into practice in the next post). 

The RIP process must be activated in the ‘config’ mode. Then we need to instruct it which interfaces should be activated in the RIP domain. This is configured in the ‘config-router’ mode (‘network’ statement).  The routers begin to ‘chat’ and advertise their routing tables every 30 seconds.

Pretend that R1’s timer of sending the advertisement has just kicked in (pic. 2). R1 is advertising its routing table out of the RIP-enabled interfaces (in my example all interfaces of all routers are in the RIP domain). This way, R2 learns about 10.1.1.0/24 subnet. So from R2’s perspective, R1 router becomes the gateway towards 10.1.1.0/24. 

Now, a word about the metric being advertised. 

Metric used in DV reflects how many routers the packet has to traverse to reach the destination network/subnet (so called 'hop-count'). R1’s routing table’s entries (subnets: 10.1.1.0/24 and 10.1.12.0/24) show the metric of ‘0’ hops (pic. 1) since they are directly connected to F0/0 and F0/1 interfaces respectively (they are local to R1). While advertising them to the neighbors (pic. 2), R1adds 1 hop (itself) to the existing metric found in the routing table.

---

NOTICE!
Bear in mind, that algorithm prompts the router to send the full routing table. Current implementation changes that behavior (split-horizon) but more on this later in the post.

---

Pic. 2 – R1’s RIP advertisements.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

R2 accepts the advertisement about 10.1.1.0/24. It puts this information in the RIP’s database and then it creates the entry in the routing table (purple color). Pay a close attention to what has just happened (pic. 2). The update arrives on R2’s F0/0 interface (RIP-enabled), sourced by the IP address of 10.1.12.1. This way, R2 considers its F0/0 the egress (outbound) interface towards the subnet advertised by R1. The IP address of the sender (10.1.12.1) becomes the next-hop IP address towards the subnet 10.1.1.0/24

Next, let’s imagine R2’s timer has expired and it is sending its routing table out F0/0 and F0/1. Please take a closer look at the picture 3 which shows this process in the graphical form. Just like previously R1 router has done, R2 is sending its routing table adding itself as an additional hop added to the existing metric (existing metric +1).

Pic. 3 – R2’s RIP advertisement.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Now, R1 and R3 accept the advertisement from R2 and register the information sent in their RIP databases (the interfaces process the update as they RIP-enabled). Appropriate entries in the routing tables also show the egress interfaces and the metric expressed in the number of ‘hops’ (how many routers the packet will have to traverse to reach the destination subnet). Also, the IP address of the gateway (the sender IP address) is registered. Again, take a look at pic.3 which shows the new entries (in purple).

Now is the time for R3 to send its own advertisement. Using the same logic you should be able to tell what is going to happen. Take a look at pic. 4 to see what is going to be advertised and what is going to be learned.

The advertisement sent out R3’s F0/0 interface is useless in our topology because there is no other router listening to it. In my next post, I will show you how to prevent a router from doing it. Advertisement sent out F0/1 interface contains information about R3’s directly connected subnet 10.1.3.0/24. Since the existing metric in R3’s routing table for this subnet is ‘0’ (directly connected to F0/0), R3 will add itself as the hop and advertise it with the metric of ‘1’ (existing metric + 1). R2is going to learn it on its F0/1 interface which becomes the outbound interface to reach the subnet 10.1.3.0/24. It is the interface to reach the advertising router’s IP address 10.1.23.3 after all.

Pic. 4 – R3’s RIP Advertisement.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Picture 4 shows this process.

This whole process of advertising the routing table out of all RIP-enabled interfaces occurs every 30 seconds but in fact, there is a jitter time introduced so this may vary between 25-30 seconds. WhenR2 advertising timer expires, it will pass the information contained in the its routing table on to R1. By doing this, R1 learns about all subnets R2 can reach, including 10.1.3.0/24 now (pic. 5).

Pic. 5 – R2’s RIP Advertisement.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

The process of spreading the information explained using this method is referred to as the ‘routing by rumor’. The state in which all routers have stable information about all networks/subnets that can be reached is called the ‘convergence’. Do not confuse it with ‘convergent networks’ which allow all sort of packet transmissions (voice, video, and data).

Take a look at picture 6. It shows that all routers can reach all the subnets available in the RIP domain. Convergence has been accomplished since their routing table are synchronized and up-to-date.

Pic. 6 – Convergence Achieved.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

The method of distributing information presented is prone to introduce two problems:

• Routing Loops
• Counting to Infinity

Of course, they have been resolved by using different techniques which I am going to explain later in the post.

Let’s take a look at the downside of using distance vector algorithm.

Routing Loops
In the picture 7, R1’s F0/0 interface. As soon as the IOS detects this fact, the entry in the routing table about 10.1.1.0/24 is immediately flushed (removed from the routing table completely).

Pic. 7 – R1’s F0/0 Interface Goes Down.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

As per the DV algorithm R1 would still wait till its advertisement timer expires. So instead of sending this ‘update’ immediately after it has lost the subnet, it will wait till its timer says: ‘now you can advertise your routing table’. This behavior might create a loop between R1 and R2 as far as the 10.1.1.0/24 subnet is concerned. Consider this situation depicted below.

Pic 8 – R2’s Advertising Timer Expires.

Icons designed by: Andrzej Szoblik - http://www.newo.pl


R2 is advertising its full routing table out of all RIP-enabled interfaces. In this announcement, there is 10.1.1.0/24 subnet. The metric being advertised is: ‘2’ (the existing metric on R2 + 1). By now, you already know that the advertising router is going to add itself as the hop to the metric of the subnet/network it advertises.

Here is the issue. R1 is receiving 10.1.1/0/24 with the metric of 2 hops, the egress interface (the one the ad came on) is F0/1, and the next-hop-address is 10.1.12.2. Look at the pic. 8 and tell me (I can’t hear you though), what would you do if you were R1? Obviously, you would reject this information because by looking at the topology diagram, you already know that 10.1.1.0/24 is inaccessible (down) now, and the only way to reach it is through R1, right?

But the problem is, that routers using DV algorithm do NOT know the topology like explained in the characteristics section. In fact, R1 IS going to accept the information and treat R2 as the gateway towards 10.1.1.0/24 !!!

Wow! As ridiculous as it sounds, it is exactly what would happen according to the rules set by the designers of this algorithm. So R1’s routing table is going to look like shown in the picture 8. Take a look at it now again!

We have a loop between R1 and R2 regarding 10.1.1.0/24. If R2 receives the packets destined to 10.1.1.0/24 subnet, according to its knowledge (current routing table), it is going to send it out F0/0 interface towards R1. This one in turn, will use its F0/1 interface for the destination 10.1.1.0/24, sending it back to R2. The packets will be looped until their TTL values are decremented reaching the value of TTL=0. Then, a router must drop the packet.

Counting To Infinity
A routing loop is not going to be the only problem here. R1 is going to accept advertisements fromR2 regarding 10.1.1.0/24 with the number of hops equal '2’. When R1 advertises its own routing table, it is going to add itself (as the hop) to the metric that already exists in the routing table. Look what is going to happen (pic. 9)

Pic. 9 – R1’s RIP Advertisement.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Initially, R2 is going to ignore the information about 10.1.1.0/24 from R1 containing the metric of ‘3’ hops since it has much better entry in the routing table (lower metric). However, it was R1 that initially sent the metric of ‘1’ hop. Now, the same R1 router keeps sending the metric of ‘3’ hops. The previous metric of ‘1’ is no longer refreshed. Since it uses the aging timer of 180 seconds (how long the information is valid), it finally accepts the entry with the metric of ‘3’ hops instead.

Then R2 begins to advertise the metric of 4 regarding 10.1.1.0/24 subnet out F0/0 and F0/1. You can predict what is going to happen. Remember, that entries must be refreshed every 30 seconds. If they are not refreshed, the ‘Invalidation Timer’ (180 seconds), allows to accept the entry with worse metric than previously. Take a look at the sequence of events in the picture 10.

Pic. 10 – Larger and Larger Metric Propagation.

Icons designed by: Andrzej Szoblik - http://www.newo.pl

It would last forever despite of the fact that 10.1.1.0/24 is not reachable at all!

The Distance Vector algorithm uses a few techniques to prevent these two problems from happening. Here they are:

• Triggered Update (aka flash update)
• Route Poisoning
• Maximum Metric (RIP considers 16 hops as inaccessible)
• Poison Reverse
• Hold-Down Timer
• Split-Horizon

These methods deserve a few words of explanation.

Triggered Update

IOS uses this method to send the update immediately rather than wait for the advertisement timer to expire. However, there is no guarantee that some router in the chain is not going to send its own information before it receives this update. This might still lead to a situation where the two problems occur. So this method, as the only solution here, is not enough to make it work. Other methods must be used as well in order to avoid routing loops and counting to infinity.

Route Poisoning

Upon losing subnet/network reachability, a router is sending a triggered update. This update is going to include the maximum metric value (RIP=16 hops) which is considered as ‘subnet/network inaccessible’ (cannot be reached).

Maximum Metric (RIP=16)

If a RIP router receives an update about a network/subnet with the metric of 16 hops it is considered as inaccessible. This way, the advertising router is excluded from the list of gateways for the subnet/network advertised with the maximum metric.

Poison Reverse

Once a router receives the advertisement including the maximum metric, if it does not have an alternate path towards the subnet/network lost, it is going to send the same subnet/network prefix with the maximum metric (RIP=16) informing the other routers about it. This will also be sent back to the sender of this information it does not have an alternate path (this might be seen as violation of split-horizon, but remember the metric is the maximum value). Poisoning the path back to the advertising router is the way of informing it that the receiver of this information has no alternate path available either.

Hold-Down Timer

Upon receiving information from a neighbor that a subnet/network is inaccessible, the receiving router is going to enable a hold-down timer for 180 seconds. During this time, the receiving router keeps sending packet to the destination being inaccessible for some time rather than withdrawing the entry from its routing table. Why?

In the past, the routers did not have that much power and the media were unreliable. Interfaces were prone to flaps more often than in today’s reliable networks. An ‘interface flap’ is the condition when it goes down and up subsequently in a very short space of time (1-2 seconds perhaps). Under such circumstances, a router would advertise network as inaccessible and then as accessible again. Since it takes some CPU power to withdraw the entry and put it back in, the designers preferred to wait a bit longer to be absolutely sure (180 seconds by default) that the entry was supposed to be removed from the routing table. In case of an interface flapping, not only would the packets still be delivered but the CPU would not waste its ‘precious’ cycles on removing and putting the entry back in the routing table. 

Split-Horizon

This method prevents the loops from occurring in the scenario we have talked about. This technique prevents a router from sending information it learned back out the interface it was received on. Consider our first example. R2 sent information about 10.1.1.0/24 before R1 had had a chance to send the maximum metric towards R2 (subnet down). Split-Horizon prevents R2 from sending information about 10.1.1.0/24 it learned on its F0/0 interface back out the same interface. As a result of that, R1 is never going to receive information it sent towards R2 (10.1.1.0/24) and believe R2 could be the gateway to 10.1.1.0/24. Thus, there is no loop

In my next post I’m going to show you how to enable RIP and how all these techniques work in practice.

Lesson 19 - Spanning-Tree Protocol Overview

Vlans described in the previous posts are very important elements of building modern networks. Equally important piece of technology is IEEE 802.1D, commonly known as Spanning-Tree Protocol. In the following few posts, I will focus on its application and basic operation.

If your network consists of layer 2 switches that allow computers connect and exchange data, you will need to consider the design that can withstand some types of failure.


Redundant Connections 

Consider the following layer 2 design. Imagine that the SW1, SW2 and SW3 switches connect many devices and there is only a single connection between the switches like depicted in the Pic1.

Pic. 1 - Switch Topology Without Redundancy

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Should either of the links between the switches break, the communication between many devices fail. Such design creates a single point of failure. We could easily tweak this simple design to make it more resilient by adding an extra path between SW2 and SW3. The below picture shows this modified design.

Pic. 2 - Redundant Paths

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Unfortunately, creating the extra path here comes at a cost. The redundant connection (Pic. 2) between SW2 and SW3 creates a loop. The loop in turn, will create three serious problems. The last one in the list will eventually render our system unavailable. Let's see what these problems are.

Duplicate Frame Delivery

Pic. 3 - Problem 1 - Duplicate Frame Delivery

Icons designed by: Andrzej Szoblik - http://www.newo.pl 

Look at the pic. 3 and imagine SW2 and SW3 do not have the MAC address of PC3(0000.3333.3333) in their databases (CAM). This can happen if the PC3 doesn't speak for more than five minutes. This is the default time MAC address is kept in the database without refreshing it. Then, we have PC1 sending frame towards PC3. As you recall, SW2 will flood the frame out of its active ports if it does not know where PC3 is located (unknown destination MAC address). The frame travels out SW2's port F0/13 towards SW1 and out the port F0/12 towards SW3. SW2 will deliver the frame to PC3. Since SW3 floods the frame out as well, it will be sent towards SW1 out of its port F0/14. Then, SW1 obediently delivers the same copy of the frame to PC3 again.


MAC Address Table Instability
Another issue caused by the loop we have created will make switches change the MAC addresses depending on where they hear the sender. Consider pic. 4 below.

Pic. 4 - Problem 2 - MAC address table instability

Icons designed by: Andrzej Szoblik - http://www.newo.pl

Again, let us assume that none of the switches in the picture knows where PC3 is connected. This means they have not learned its MAC address yet. In our scenario, PC1 sends the frame to PC3(destination MAC: 0000.3333.3333). SW2 floods the frame out F0/12 and F0/13 ports.

Now, SW3 receives this frame sourced with 0000.1111.1111 MAC address (PC1). It learns the source MAC address and maps it to its F0/12 port where it arrived. Since SW1 does not know wherePC3 is connected (at least right now) it will flood this frame out all active ports. This way, the frame is sent out SW1's port F0/14 towards SW3. SW3, upon receiving the frame on its F0/14 port, reads the source MAC address (0000.1111.1111) and maps it to port F0/14 this time. This causes a little confusion as SW3 learned it earlier on and it was port F0/12 before. Previous mapping is removed and F0/14 becomes the outbound port for 0000.1111.1111 now.

Broadcast Storm
The last problem is really severe. It can bring our traffic to a halt. Take a look at pic. 5 below.

Pic. 5 - Problem 3 - Broadcast Storm

Icons designed by: Andrzej Szoblik - http://www.newo.pl

In this scenario, PC1 sends a broadcast frame. SW2 upon receiving it, floods it out all its active ports. SW1 receives it on port F0/13 and floods it out of other ports. SW3 receives the broadcast frame on its F0/12 port and floods it. Then, a tad later it receives this same broadcast frame from SW1 and again it floods it out all active ports except the port it arrived on. You can write the rest of the story on your own. This broadcast is running in the loop in both directions endlessly. Well, not exactly endlessly. It is true that there is not mechanism to stop it, but all three switches in the topology will be so busy sending out this broadcast, that eventually all its resources are consumed and they stop sending anything at all. If you look at switches that experience a broadcast storm, you will notice that all their LEDs are flashing amber like a Christmas tree. In a few seconds the switches become unresponsive. An attempt to access them remotely using SSH/telnet will fail. Even console connection may refuse to accept your commands. The only way to bring the switches back to the operation is to break the loop by pulling one of those cables.

So, what can we not have redundancy in our layer 2 topology? Of course, we can.

We will run Spanning-Tree Protocol (turned on by default), which will dynamically block redundant connections creating a loop free topology. Should the primary link fail, the one that is in the blocking state will start forwarding the traffic in about 30 seconds by default. Of course, we will need something much faster than 30 seconds, but I will show you that as soon as we know how STP works.

Here I am going to give you just an overview of its operation. But the devil is in the details which we will scrutinize in my next post.

Spanning-Tree Protocol Overview
STP is a layer 2 loop prevention mechanism. Switches running this protocol use special frames called Bridge Protocol Data Unit (BPDU). These frames contain enough information to allow the switches to create a loop free topology. This magic is accomplished using three distinct phases:

1. Elect a single switch to be the root bridge machine which is the central device in the layer 2 network. This machine will have all its ports in the forwarding state (designated port role).
2. All other switches (non-root switches), will select a single path towards the root bridge. That port is called the 'root port' and will be forwarding traffic that is destined out of the switch through the root bridge. This path is the least cost (best) path towards the root.
3. All other switches will select a single path per segment in order to block stop the loop. The port that is forwarding traffic is called designated port. The port that is blocking traffic to stop the loop is called non-designated port.

I will explain all the terms and the above process in details in my next post. Meanwhile, check the pic. 6 first.

Pic. 6 - Spanning-Tree Protocol

Icons designed by: Andrzej Szoblik - http://www.newo.pl

In the above picture, SW1 has been elected as the root bridge. SW2 uses port F0/13 as its root port (the best, or the least cost path towards the root). SW3 uses it port F0/14 as the root port. SW3blocks the port F0/12 to stop the loop. SW2 keeps sending BPDU frames originated by the root bridge (SW1) out its F0/12 port towards SW3.

Now, what is really fascinating that the loop free structure like the above is done automatically (although you want and will affect how it works), and the fact that if the communication betweenSW2 and SW1, or SW3 and SW1 is broken, the SW3 port F0/12 will be put in the forwarding state.

If you are interested in the details how STP works please read my next post (lesson 20).