Switch Configurations

LAN switches vary in their physical design. Currently, there are three popular configurations in use:

• Shared memory - This type of switch stores all incoming packets in a common memory buffer shared by all the switch ports (input/output connections), then sends them out via the correct port for the destination node.


• Matrix - This type of switch has an internal grid with the input ports and the output ports crossing each other. When a packet is detected on an input port, the MAC address is compared to the lookup table to find the appropriate output port. The switch then makes a connection on the grid where these two ports intersect.


• Bus architecture - Instead of a grid, an internal transmission path (common bus) is shared by all of the ports using TDMA. A switch based on this configuration has a dedicated memory buffer for each port, as well as an ASIC to control the internal bus access.

Transparent Bridging

Most Ethernet LAN switches use a very cool system called transparent bridging to create their address lookup tables. Transparent bridging is a technology that allows a switch to learn everything it needs to know about the location of nodes on the network without the network administrator having to do anything. Transparent bridging has five parts:

• Learning


• Flooding


• Filtering


• Forwarding


• Aging

Here's how it works:


Click on the menu terms to learn more about how transparent
bridging works.

In the next section, you'll get a step-by-step description of how transparent bridging works.

Transparent Bridging: The Process

Here's a step-by-step description of transparent bridging:

• The switch is added to the network, and the various segments are plugged into the switch's ports.


• A computer (Node A) on the first segment (Segment A) sends data to a computer (Node B) on another segment (Segment C).


• The switch gets the first packet of data from Node A. It reads the MAC address and saves it to the lookup table for Segment A. The switch now knows where to find Node A anytime a packet is addressed to it. This process is called learning.


• Since the switch does not know where Node B is, it sends the packet to all of the segments except the one that it arrived on (Segment A). When a switch sends a packet out to all segments to find a specific node, it is called flooding.


• Node B gets the packet and sends a packet back to Node A in acknowledgement.


• The packet from Node B arrives at the switch. Now the switch can add the MAC address of Node B to the lookup table for Segment C. Since the switch already knows the address of Node A, it sends the packet directly to it. Because Node A is on a different segment than Node B, the switch must connect the two segments to send the packet. This is known as forwarding.


• The next packet from Node A to Node B arrives at the switch. The switch now has the address of Node B, too, so it forwards the packet directly to Node B.


• Node C sends information to the switch for Node A. The switch looks at the MAC address for Node C and adds it to the lookup table for Segment A. The switch already has the address for Node A and determines that both nodes are on the same segment, so it does not need to connect Segment A to another segment for the data to travel from Node C to Node A. Therefore, the switch will ignore packets traveling between nodes on the same segment. This is filtering.


• Learning and flooding continue as the switch adds nodes to the lookup tables. Most switches have plenty of memory in a switch for maintaining the lookup tables; but to optimize the use of this memory, they still remove older information so that the switch doesn't waste time searching through stale addresses. To do this, switches use a technique called aging. Basically, when an entry is added to the lookup table for a node, it is given a timestamp. Each time a packet is received from a node, the timestamp is updated. The switch has a user-configurable timer that erases the entry after a certain amount of time with no activity from that node. This frees up valuable memory resources for other entries. As you can see, transparent bridging is a great and essentially maintenance-free way to add and manage all the information a switch needs to do its job!

In our example, two nodes share segment A, while the switch creates independent segments for Node B and Node D. In an ideal LAN-switched network, every node would have its own segment. This would eliminate the possibility of collisions and also the need for filtering.

Redundancy

When we talked about bus and ring networks earlier, one issue was the possibility of a single point of failure. In a star or star-bus network, the point with the most potential for bringing all or part of the network down is the switch or hub. Look at the example below:







In this example, if either switch A or C fails, then the nodes connected to that particular switch are affected, but nodes at the other two switches can still communicate. However, if switch B fails, then the entire network is brought down. What if we add another segment to our network connecting switches A and C?







In this case, even if one of the switches fails, the network will continue. This provides redundancy, effectively eliminating the single point of failure.

But now we have a new problem.

Broadcast Storms

In the last section, you discovered how switches learn where the nodes are located. With all of the switches now connected in a loop, a packet from a node could quite possibly come to a switch from two different segments. For example, imagine that Node B is connected to Switch A, and needs to communicate with Node A on Segment B. Switch A does not know who Node A is, so it floods the packet.







The packet travels via Segment A or Segment C to the other two switches (B and C). Switch B will add Node B to the lookup table it maintains for Segment A, while Switch C will add it to the lookup table for Segment C. If neither switch has learned the address for Node A yet, they will flood Segment B looking for Node A. Each switch will take the packet sent by the other switch and flood it back out again immediately, since they still don't know who Node A is. Switch A will receive the packet from each segment and flood it back out on the other segment. This causes a broadcast storm as the packets are broadcast, received and rebroadcast by each switch, resulting in potentially severe network congestion.

Which brings us to spanning trees...

Spanning Trees

To prevent broadcast storms and other unwanted side effects of looping, Digital Equipment Corporation created the spanning-tree protocol (STP), which has been standardized as the 802.1d specification by the Institute of Electrical and Electronic Engineers (IEEE). Essentially, a spanning tree uses the spanning-tree algorithm (STA), which senses that the switch has more than one way to communicate with a node, determines which way is best and blocks out the other path(s). The cool thing is that it keeps track of the other path(s), just in case the primary path is unavailable.

Here's how STP works:

• Each switch is assigned a group of IDs, one for the switch itself and one for each port on the switch. The switch's identifier, called the bridge ID (BID), is 8 bytes long and contains a bridge priority (2 bytes) along with one of the switch's MAC addresses (6 bytes). Each port ID is 16 bits long with two parts: a 6-bit priority setting and a 10-bit port number.


• A path cost value is given to each port. The cost is typically based on a guideline established as part of 802.1d. According to the original specification, cost is 1,000 Mbps (1 gigabit per second) divided by the bandwidth of the segment connected to the port. Therefore, a 10 Mbps connection would have a cost of (1,000/10) 100.
  
  To compensate for the speed of networks increasing beyond the gigabit range, the standard cost has been slightly modified. The new cost values are:
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  

  Bandwidth	STP Cost Value
  4 Mbps	250
  10 Mbps	100
  16 Mbps	62
  45 Mbps	39
  100 Mbps	19
  155 Mbps	14
  622 Mbps	6
  1 Gbps	4
  10 Gbps	2

  
  You should also note that the path cost can be an arbitrary value assigned by the network administrator, instead of one of the standard cost values.


• Each switch begins a discovery process to choose which network paths it should use for each segment. This information is shared between all the switches by way of special network frames called bridge protocol data units (BPDU). The parts of a BPDU are:
  
  
  
  • Root BID - This is the BID of the current root bridge.
  
  
  • Path cost to root bridge - This determines how far away the root bridge is. For example, if the data has to travel over three 100-Mbps segments to reach the root bridge, then the cost is (19 + 19 + 0) 38. The segment attached to the root bridge will normally have a path cost of zero.
  
  
  • Sender BID - This is the BID of the switch that sends the BPDU.
  
  
  • Port ID - This is the actual port on the switch that the BPDU was sent from.
  
  
  
  All of the switches are constantly sending BPDUs to each other, trying to determine the best path between various segments. When a switch receives a BPDU (from another switch) that is better than the one it is broadcasting for the same segment, it will stop broadcasting its BPDU out that segment. Instead, it will store the other switch's BPDU for reference and for broadcasting out to inferior segments, such as those that are farther away from the root bridge.


• A root bridge is chosen based on the results of the BPDU process between the switches. Initially, every switch considers itself the root bridge. When a switch first powers up on the network, it sends out a BPDU with its own BID as the root BID. When the other switches receive the BPDU, they compare the BID to the one they already have stored as the root BID. If the new root BID has a lower value, they replace the saved one. But if the saved root BID is lower, a BPDU is sent to the new switch with this BID as the root BID. When the new switch receives the BPDU, it realizes that it is not the root bridge and replaces the root BID in its table with the one it just received. The result is that the switch that has the lowest BID is elected by the other switches as the root bridge.


• Based on the location of the root bridge, the other switches determine which of their ports has the lowest path cost to the root bridge. These ports are called root ports, and each switch (other than the current root bridge) must have one.


• The switches determine who will have designated ports. A designated port is the connection used to send and receive packets on a specific segment. By having only one designated port per segment, all looping issues are resolved!
  
  Designated ports are selected based on the lowest path cost to the root bridge for a segment. Since the root bridge will have a path cost of "0," any ports on it that are connected to segments will become designated ports. For the other switches, the path cost is compared for a given segment. If one port is determined to have a lower path cost, it becomes the designated port for that segment. If two or more ports have the same path cost, then the switch with the lowest BID is chosen.


• Once the designated port for a network segment has been chosen, any other ports that connect to that segment become non-designated ports. They block network traffic from taking that path so it can only access that segment through the designated port.

Each switch has a table of BPDUs that it continually updates. The network is now configured as a single spanning tree, with the root bridge as the trunk and all the other switches as branches. Each switch communicates with the root bridge through the root ports, and with each segment through the designated ports, thereby maintaining a loop-free network. In the event that the root bridge begins to fail or have network problems, STP allows the other switches to immediately reconfigure the network with another switch acting as root bridge. This amazing process gives a company the ability to have a complex network that is fault-tolerant and yet fairly easy to maintain.