Spanning Tree Protocol
Earlier, we mentioned that one of the functions of a switch was Layer 2 Loop removal. This is a critical feature, as without it many switched networks would completely cease to function. Either accidentally or deliberately in the process of creating a redundant network, the problem arises when we create a looped switched path. A loop can be defined as two or more switches that are interconnected by two or more physical links.
Switching loops create three major problems:
- Broadcast Storms—Switches must flood broadcasts, so a looped topology will create multiple copies of a single broadcast and perpetually cycle them through the loop.
- MAC table instability—Loops make it appear that a single MAC address is reachable on multiple ports of a switch, and the switch is constantly updating the MAC table.
- Duplicate frames—Because there are multiple paths to a single MAC, it is possible that a frame could be duplicated to be flooded out all paths to a single destination MAC.
All these problems are serious and will bring a network to an effective standstill unless prevented.
Figure 12.1 illustrates a looped configuration causing a broadcast storm:
Figure 12.1 A Layer 2 (switching) loop.
Other than simple error, the most common reason that loops are created is because we want to build a redundant or fault-tolerant network. By definition, redundancy means that we have a backup, separate path for data to follow in the event the first one fails. The problem is that unless the backup path is physically disabled—perhaps by unplugging it—the path creates a loop and causes the problems mentioned previously. We like redundant systems; we do not like loops and the problems they cause. We need a mechanism that automatically detects and prevents loops so that we can build the fault-tolerant physical links and have them become active only when needed. The mechanism is called the Spanning Tree Protocol (STP). STP is a protocol that runs on bridges and switches to find and block redundant looped paths during normal operation. Spanning Tree was originally developed by the Digital Equipment Corporation (DEC), and the idea was adopted and modified by the IEEE to become 802.1d. The two are incompatible, but it is exceedingly rare to find a DEC bridge these days, so the incompatibility is not usually a problem.
STP's basic function is to create a loop-free path to a root bridge. The root bridge is the bridge or switch that is the root of the Spanning Tree, with the branches being loop-free paths to the other switches in the system. The Root is the switch with the lowest Bridge ID; the ID is determined by a combination of an administrative Priority and the MAC address of the switch. The Priority is set to 32,768 (8000 hex) by default; if we leave the Priority at the default, whatever switch has the lowest MAC will be the Root. Figure 12.2 illustrates a simple Root selection when all switches are using the default Priority.
Figure 12.2 Root Bridge Selection with the default Priority.
We cannot change the MAC address of a switch, so what happens if Switch A in the previous example happens to be an old, slow Catalyst 1900? It might get elected the Root because it has a low MAC address, but we really don't want it to be the Root: Usually, we would choose a big, fast switch at the core of the network as the Root. Let's say that Switch C is a hot new switch and we want it to be our Root; how do we override the existing election? The answer is to change the default Priority—remember, the lowest ID wins the election, and the ID is the Priority prepended to the MAC. The ID is one long string, so lowering the Priority makes the ID lower. Thus, if we change the Priority of Switch C to a low value, it will win the election despite the fact that it has a higher MAC than A. Figure 12.3 illustrates this.
Figure 12.3 Root Election with a modified Priority.
STP Communication with BPDUs
To determine the presence of loops and to block loops, switches must be capable of communicating with each other about the various connections they have. This communication in STP is carried out by the exchange of Bridge Protocol Data Units (BPDUs). The 802.1d BPDU is multicasted every two seconds and includes information the switches need to decide if there are loops, how to fix them, and which switch is the Root. Figure 12.4 shows the fields in an 802.1d BPDU; note the fields for the Bridge ID, the Root ID, and the Root path cost.
Figure 12.4 Detailed contents of 802.1d BPDU packet.
STP assigns different ports on a switch as different types, depending on where the Root is and where the loops are in the topology. The sections that follow describe the port types and how they are selected.
The Root port on a switch is the one port that has the lowest cost path to the Root switch. Path cost is calculated based on the bandwidth of the links. Table 12.1 lists the IEEE-defined values for STP path cost; note that there are old and new values. The new values were defined because of the increasingly widespread availability of multi-Gigabit link speeds; previously, a 1Gbs link had the same cost as a 10Gbps link. That made no sense and would create suboptimal STP topologies, so the costs were revised.
Table 12.1. STP Path Costs, Old and New
New Cost Value
Old Cost Value
After the switches have elected the Root for the system, each switch must then decide which port it will use to reach the Root. Some switches will have only one port that can reach the Root at all; some might have several, depending on the number and location of uplinks between the switches in the system. The exchange of BPDUs that decides the Root election also tells each switch about the path costs to reach the Root (as indicated by the value of the Root Path Cost field in the BPDU). Each switch adds its own path cost to the path cost received from the neighboring switch and chooses the port with the lowest cost as the Root Port. Figure 12.5 illustrates root port selection in a simple switched network.
Figure 12.5 The Root port is the one with the lowest Root Path Cost.
Note that the Root itself does not have any Root ports: It does not need to reach the Root—it is the Root!
For each LAN segment, there must be one Designated port. This is the port that will forward traffic to the Root from the LAN segment. The Designated port is the port that has the least cost path to the Root from the LAN segment.
The Root switch has only Designated ports. Because it is the root, it won't have a Root port, and it can't block any of the ports that connect to other switches (because that would make the other switch's Root ports not work).
In Figure 12.6, our three switches have already elected the Root and chosen their Root ports. Switch A is the Root, so all of its ports are Designated. Switches B and C must next choose which port will block and which port will be designated on the link between them.
Figure 12.6 The Designated port selection process.
The first criterion examined is which switch has the lowest root path cost. In our setup here, B and C each connect to the Root with a 100Mbps connection, with an STP cost of 19. By examining each other's BPDUs, B and C realize that they are tied for root path cost.
This is a very common scenario in modern networks where switches are directly connected over full-duplex crossover cables. One of the switches must block its port to stop the loop. The second criterion (the first tiebreaker) is the lower Bridge ID: in this case, Switch B wins and Switch C must block its port.
As we get into more complex switched systems, we get into situations where additional criteria (tiebreakers) are needed. The full list is examined in the next section, "Port Type Selection."
Port Type Selection
The order of criteria a switch goes through when deciding its Root and Designated ports is as follows:
- The port with the lowest cumulative Root Path Cost will be the Root port/Designated port.
- If tied between multiple ports, the port that connects to the neighboring switch with the lowest Bridge ID becomes the Root port/Designated port.
- If there are multiple connections to that same switch, the port with the lowest assigned STP priority will be the Root port/Designated port.
- If tied, the port with the lowest hardware number (Fa0/1 is lower than Fa0/2) will be the Root port/Designated port.
A Blocked port is neither the Root port nor the Designated port, but is part of the redundant links between switches. In other words, it lost in the election to choose the active Root or Designated ports, but it might take over one of these roles if the active port failed. A Blocked port is the one that actually stops the loop, so it is just as important as the Root or Designated. A Blocked port does not send data; it only receives BPDUs.
Convergence is the term used to describe the process STP goes through to achieve a stable, loop-free network. (The same term is used with reference to routing information stability as well.) When all switches have elected the Root and decided on their Root, Designated, and Blocked ports, the system is said to be converged.
With 802.1d STP, each port on each switch goes through four distinct port states in the process of convergence:
Blocking—When a switch boots up, all ports start in the blocking state. This is to prevent loops during the time that the STP topology is converging. A port that is a link between switches will stay blocked unless it becomes a Root or Designated port. Blocked ports send no data at all (not even BPDUs), but they do listen for (receive) BPDUs from other switches. All ports will also go to Blocking mode if a Topology Change Notification (TCN) BPDU is received. TCNs are issued when a new link is added or removed—the topology of the switched system is altered. When this happens, STP reacts by blocking all ports until loop-free convergence is achieved.
If a switch dies or a link between switches fails, the other switches connected to it wait for a specific time until they begin the STP convergence process. This interval is called the Max Age Timer, and by default it is 20 seconds. Effectively, it means that a switch will wait until it has missed 10 BPDUs (which are sent every 2 seconds) from a connected switch before it kicks in the STP recalculation.
- Listening—The Listening state enables a Blocked port to begin sending its own BPDUs. By default, the Listening state is 15 seconds.
- Learning—The Learning state is when the switch begins populating its MAC address table. It is not yet forwarding any frames, but it is getting ready to forward by building as complete a MAC table as it can. The Listening state is also 15 seconds by default. The Listening and Learning states together are called the Forward Delay, and you might see their two 15-second timers represented as a single 30-second timer called the Forward Delay Timer.
- Forwarding—The Forwarding state, as its name implies, is when the port starts forwarding frames. This is simply normal operation for a port that is not blocked.
If you take a quick look at these states and their timers, you can see that in 802.1d STP, reaching convergence can take anywhere from 30 to 50 seconds (Forward Delay [15+15]+ MaxAge = 50 seconds). Understand that during this 30 to 50 seconds, no frames are being forwarded at all—no data is being sent anywhere because every port on every switch is either Blocking, Listening, or Learning. This is, of course, very detrimental to the productivity and utility of a network, especially a modern, busy one. A 50-second delay every time a topology change happens is unacceptable, so Cisco (and then the IEEE) created several enhancements to 802.1d STP to speed up the process of convergence. Some of these enhancements are discussed in the following section.
The Rapid Spanning Tree protocol (RSTP, IEEE 802.1w—remember, 802.1w is Wapid Spanning Twee) has many of its roots in Cisco-created enhancements to ordinary 802.1d STP. The primary goal of these enhancements is to speed up convergence. There are no timers in RSTP; instead, the BPDU becomes much more detailed and informative so that switches can gather more information with greater accuracy. New port states have been defined as shown in Table 12.2.
Table 12.2. RSTP Port States
Switches wait for only three missing BPDUs before commencing the Spanning-Tree recalculation process. The process of convergence is itself much more rapid because new port types have been defined as well. In addition to the Root and Designated port types in STP, RSTP defines the Alternate and Backup port types. The Alternate port is the port that will become the Root port if the primary Root port fails. The Backup port is the port that will become the Designated port if the primary Designated port fails. The BPDUs in RSTP convey information about these port types to neighboring switches. This enhanced communication allows for quicker convergence, without relying on the 30–50 second timers in STP.
Another significant improvement in convergence speed comes from the Rapid Transition to Forwarding (RTF) features of Edge ports and link types. Edge ports are ports that are connected to non–STP-capable devices such as PCs, servers, or routers. These devices will not normally create STP loops, so there is no need for them to block to prevent loops. This function is enabled by Cisco's portfast command feature. With PortFast configured, a switch port will stop sending BPDUs (after a few have been sent as a precaution to prevent loops) and transition to the forwarding state almost immediately. This is very useful to get frames moving through the switch so hosts can get on with business—picture a database server and a PC connected to the same switch; they would not have to wait the 50 seconds for STP convergence if portfast was configured on both ports. In addition, if a port configured for portfast does receive a BPDU (perhaps because someone plugged a switch in), by default it will disable portfast and start STP on that port to prevent loops. You can also optionally configure the switch port with BPDU Guard to shut down the port if it receives a BPDU. This is more secure because it prevents the unauthorized installation of switches. BPDU Guard is covered in the Cisco CCNP curriculum.
The interface-configuration syntax to configure a Catalyst switch port with Portfast looks like this:
Or, to set all ports to use portfast by default, use the global configuration command:
Switch(config)#spanning-tree portfast default
To turn PortFast off, use the spanning-tree portfast disable interface configuration command.
Another Cisco enhancement deals with port security; this feature set allows you (among several other options) to disable a port if more than one MAC address is detected as being connected to the port. This feature is commonly applied to ports that connect security-sensitive devices such as servers.
The following command syntax restricts access to a single MAC address and shuts the port down if another MAC connects:
Switch(config)#interface fa0/21 Switch(config-if)#switchport mode access Switch(config-if)#switchport port-security Switch(config-if)#switchport port-security maximum 1 Switch(config-if)#switchport port-security violation shutdown
Link Types refers to a port setting of either full duplex or half duplex. If a port is set for full duplex, RSTP assumes that it is a candidate for rapid transition because there can be only one other device at the end of such a connection. If it is set for half duplex, however, it is conceivable that there could be multiple STP-capable devices on that segment, so by default the RTF functions are disabled. It is possible to override this default.