January/February 2011

Understanding Ethernet switches and routers

By George Thomas

Editorial note: This is the first of a two-part article on Ethernet switches and routers. The first part deals with switches, and the second will focus on routers. The second part will run as a web exclusive on InTech's website (www.isa.org/link/Basics_main).

When you go to a computer store to purchase a device that will access the Internet, they will try to sell you a router. In most instances, the router will come with a built-in switch (shortened term for switching hub) so you can connect several Ethernet devices to just one device. So what is the difference between an Ethernet router and an Ethernet switch? The long answer to that question requires an examination of the Open Systems Interconnection Model, which is frequently used to explain how communication networks operate.

In the seven-layer model, each layer provides a unique service. Communication between two stations begins at the Application layer with the sending station initiating a message to a receiving station via a common medium. With respect to this model, Ethernet provides services at the Physical and Data Link layers through the use of bridges and repeaters. An Ethernet switch is classified as a bridge and therefore operates at the data link layer while routers operate at the Network layer. Let's try to understand why.

The lowest layer is the physical layer that defines the basic signaling on the medium. Ethernet transmits symbols representing logic "ones" or "zeros" across the medium to another station that decodes the symbols to extract the data. Although Ethernet will operate with coaxial cable as the medium, modern Ethernet networks incorporate twisted-pair cabling. If the path is too long, a repeater can be used to extend distance. If fiber optic cable is preferred, media converters can be used. If multiple devices need to share the connection, a repeating hub (commonly called just a hub) is used. All three of these devices reside at the physical layer because they do nothing more than process symbols on the medium.

One layer above the physical layer is the data link layer. Ethernet is a local area network technology with end stations assigned unique 48-bit addresses. These addresses are called media access control (MAC) addresses. When data is to be sent from one Ethernet station to another, the data is first arranged in frames. The destination and source addresses are appended so the intended station knows it is to receive the message and who sent it. Other parts of the frame include the Preamble, which alerts the receiving station a frame is coming, a Type or Length field that identifies either the type of data or length of the data field, the Data field itself, and the Frame Check Sequence used to verify the integrity of the frame. The payload of the frame is the actual data. Everything else is overhead.

 understand 1

Carrier Sense Multiple Access with Collision Detection (CSMA/CD)

As an Ethernet station prepares to send a frame, it first listens to the medium to verify a clear channel exists. If silence is sensed, it transmits its message and then waits a determined amount of time, called the slot time. The slot time is used for detecting a collision due to another station transmitting at the same time. If no collision is sensed, the sender assumes a successful transmission. If a collision has occurred, the sender refrains from transmitting again for an amount of time based on a backoff algorithm that incorporates randomness. An early criticism of Ethernet was this probabilistic approach to media access control was not conducive to real-time systems. There are other issues with CSMA/CD:

  • All CSMA/CD stations must reside within one collision domain to ensure all stations will detect a collision between stations located at the farthest points. This limits the geographic distance of the Ethernet network.
  • Stations residing in the same collision domain can detect all transmissions but only receive those addressed to them. All stations can transmit but not at the same time. This is called half-duplex operation or Shared Ethernet.

 understand 2

Breaking up collision domains for higher performance

A switching hub was introduced to avoid the problems of Shared Ethernet. A switching hub is much different from a repeating hub. A port on a switching hub appears to an end station as another end station except it does not consume a MAC address. To an attached end station, the switch port appears as the only other station within the collision domain. This is how it works.

Assume station A is on port 1 of an eight-port switch and station B is on port 2. Station A sends a message to station B. Switch port 1 reads the entire frame into its internal input buffer and forwards it to port 2's output buffer, which then transmits the entire frame to station B. So what is the advantage?

  • The switch has effectively created two collision domains-each appearing as a two-station link. With only two stations, collisions can be avoided altogether by creating a full-duplex link, which potentially doubles throughput.
  • With a full-duplex link, there is no collision domain, thus distance is limited only by cable losses. Fiber optic distances are no longer limited by the collision domain and can be much greater than Shared Ethernet lengths. Without a concern for a common collision domain, switches can be cascaded at will.
  • With separate collision domains on each port, each port can operate at different data rates, allowing for the mixing of data rates within the same switch.

understand 3

Another advantage to using a switch is its ability for simultaneous messages within its switch fabric. When a transmission is received on a particular switch port, the source MAC address of the sender is stored in the database of the switch. Using this Learning Process, the switch determines on what port a station can be reached.

Assume station A sends a message to station C which has been attached to port 3, but the switch does not know how to reach station C (refer to "Station A,B,C" figure). The switch will "flood" the same message to all ports. When station C eventually replies, the switch will learn that station C is on port 3 so future flooding will not be necessary. Now station A sends a message to station B, but the switch already knows station B can be reached on port 2 by using a "lookup" process so only port 2 will transmit the message. Other ports do not need to pass the message because it was only directed to station B. This frees up other ports to pass unrelated messages without a concern for stations A and B's traffic. This improves throughput over Shared Ethernet, which requires only one message can pass through a hub at any one time.

Now assume the cable on port 1 is moved to port 4. If station A does not initiate a transmission, the switch will still believe station A can be reached on port 1. For this reason, all learned addresses must be "aged" by clearing out the database periodically. Eventually, the station A - port 1 pairing will be cleared, and transmissions intended for station A will be flooded to all ports, including port 4. Station A will now receive the flooded message, and its response will allow the switch to learn its new location.

Modern switches have two more interesting features-Auto-negotiation and Auto-MDIX. With Auto-negotiation, the data rate and duplex for link partners is negotiated during initial connection. If the end station and the switch port can operate at either 10 Mbps or 100 Mbps at either half- or full-duplex, the negotiation process will select higher performing 100 Mbps full-duplex. With Auto-MDIX, either a straight-through or crossover cable can be used between an end station or switch port or between two switch ports.

ABOUT THE AUTHOR

George Thomas is president of Contemporary Controls in Downers Grove, Ill. He is a senior ISA member.