Design Guidelines

Fast Ethernet technology is the high speed successor to 10 Base-T. This makes it a very attractive technology to use for upgrading existing 10 Mbps Ethernet networks, as it is based on a familiar technology. However, the high speed imposes limits on network design which must be carefully followed to ensure a successful implementation.

Fast Ethernet Types & Cable Requirements

Fast Ethernet cabling requirements vary with the specific type of Fast Ethernet used. The most common variation, 100 Base-TX, uses Category 5 Unshielded Twisted Pair (UTP) cable. All cable runs must fully meet the Category 5 performance specifications, and should be tested to ensure compliance with the applicable standards.

The other variations of Fast Ethernet are 100 Base-T4, which uses Four Pair Category 3 (or higher) UTP, and 100 Base-FX, which uses multimode fiber optic cable.

Currently, 100 Base-T4 products are not commonly available, so this document will not address the configuration of 100 Base-T4 networks in depth. When 100 Base-T4 products are more widely available, this document will be updated to include more information on designing 100 Base-T4 networks.

Basic Operation

At the MAC (Media Access Control) layer, which controls who transmits data to the network and when they can do it, Fast Ethernet uses the same CSMA/CD protocol that 10 Mbps Ethernet uses. CSMA/CD stands for &QUOT Carrier Sense Multiple Access with Collision Detection" . The way it works is that any device needing to transmit data to the network must first wait and listen to see if anyone else is transmitting. If the network is clear, then the device can transmit a packet to the network. On the other hand, if the network is not clear, then the device needing to transmit must wait until the transmission in progress ends before starting to send its data.

It is possible in this protocol to have two devices with transmissions pending at the same time to see that the network is clear and both to begin transmitting. Naturally, this results in both transmissions being garbled, and is called a collision. All versions of Ethernet are designed to detect when this happens, and when it does the transmitting devices stop transmitting, wait a random amount of time, and try again. Since the delay is variable, it is unlikely for both devices to attempt transmitting again at the same time (one will have to wait longer than the other).

In order for the CSMA/CD protocol to work properly, we have to ensure that the worst-case round-trip signal delay between any two points on the network is not so long that a device can finish transmitting before it can detect any collisions which may occur during the transmission. There is a small delay in the cabling, hubs, and NIC cards which has to be watched out for. This delay is called Propagation Delay and, if not watched out for, can cause a complete breakdown of the CSMA/CD protocol. This will result in a slow, unreliable network.

The Ethernet standards specify that the shortest transmission unit (packet) allowed on an Ethernet network is 512 bits. Therefore, the delay introduced by the network must be less than the time required to transmit 512 bits. In a traditional 10 Mbps Ethernet, this time is fairly long by computer standards, and can usually be discounted. When we take the network to 100 Mbps, however, this time is only 10% of what it is in a 10 Mbps network, and becomes far more important to network design.

Determining Propagation Delay

Since propagation delay is vitaly important to the proper implementation of a Fast Ethernet network, it is necessary to plan the network around it. The first step is to locate the two nodes in the network which are the most distant from one another. Once this is known, the next step is to determine where the network's hub or hubs will be located.

Every device or cable run that the Fast Ethernet signal must pass through between the two most distant nodes has a Propagation Delay associated with it. This is a measurement of how much the device or cable delays the signal.

The Propagation Delay is normally measured in a unit called a Bit Time. One Bit Time is defined as the duration of one data bit on the network, in this case 1/100,000,000 second. Since the CSMA/CD protocol requires that the first bit of any transmission reaches the most distant part of the network before the last bit of the transmission is sent, and since the shortest transmission allowed is 512 bits, we must design the network such that the absolute worst case delay is less than 512 Bit Times.

One other constraint must be accounted for in the design of a Fast Ethernet network - signal strength. Due to the high frequencies involved in 100 Mbps communications, the maximum distance of any particular cable run is limited to 100 Meters of Category 5 cable, or 2000 Meters of multimode fiber. These distances are the maximum achievable under ideal conditions. It is highly likely that propagation delay concerns will limit these distances considerably.

The above information may sound very difficult and technical, however using it is actually very easy. All we do is determine what is between the two most distant nodes, look up the Propagation Delay for each device and cable run, add it all up, and see if the total is less than 512 Bit Times. If it is, and there are no cable runs exceeding 100 Meters of Category 5 or 2000 Meters of fiber, then the network is legal and should work fine. If the total Propagation Delay is greater than 512 Bit Times, then we need to segment the network with a local bridge or Fast Ethernet switch.

Propagation Delay Tables

The following table lists common Fast Ethernet devices and cable types, and the Propagation Delay associated with each:

Round-Trip Propagation Delays (in Bit Times)
Device	Delay per Meter	Maximum Delay
Two TX/FX DTEs	N/A	100
Two T4 DTEs	N/A	138
One T4 and One TX/FX DTE	N/A	127
Category 3 Cable Segment	1.14	114 (100 Meters)
Category 4 Cable Segment	1.14	114 (100 Meters)
Category 5 Cable Segment	1.112	111.2 (100 Meters)
Shielded Twisted Pair (IBM Type 1)	1.112	111.2 (100 Meters)
Fiber Optic Cable	1.00	412 (412 Meters)
Class I Repeater	N/A	140
Class II Repeater - All ports TX/FX	N/A	92
Class II Repeater - Any port T4	N/A	67
100 Base-TX to 100 Base-FX Converter	N/A	50 - 100

Table 1
Propagation Delays

Sample Network #1

Here is a very basic 100 Base-TX network. Let's determine if this configuration will work.

Figure One
Sample 100 Base-TX Network

Analysis

The network illustrated in Figure One is a very basic 100 Base-TX setup. It consists of two 100 Base-TX DTEs (the PCs), two 100 meter Category 5 cable runs, and a Class II 100 Base-TX hub. Let's determine the propagation delay of the network and see if it is less than 512 bit times:

Step One: Determine delay of each part

Two 100 Base-TX DTEs - 100 bit times
100 Meter Cat-5 cable run - 111.2 bit times
100 Meter Cat-5 cable run - 111.2 bit times
Class II Repeater (All TX) - 92 bit times

Step Two: Add the delays of all parts together

100 + 111.2 + 111.2 + 92 = 414.4

Step Three: Subtract the result from 512:

512 - 414.4 = 97.6

Result

Since the result of Step Three resulted in a positive number, the network is valid and will work without any problems.

Sample Network #2

Here is a more complicated network. Let's see if this one is legal:

Figure Two
Second Sample 100 Base-TX Network

Analysis

This network is a more complex example than the first one we looked at. In this network, we have three DTEs, two Class II repeaters (hubs), and four Category 5 cable runs. Let's look at it step by step.

Step One: Determine Worst-case path.

Look at Figure Two and find the two DTEs which are the farthest apart from one another. The way to determine which ones are the farthest apart is by counting the number of hubs and cable runs between every combination of DTEs, and picking the ones with the highest count. In this case, the two DTEs which are most distant are PC 2 and PC 3.

Step Two: Determine what's between worst-case DTEs

Between PC 2 and PC 3 we have the following:

Two 100 Base-TX NIC cards, Two Class II Hubs, an 80-meter Category 5 cable run, a 5 meter Category 5 cable run, and a 100 meter Category 5 cable run.

Step 3: Determine round-trip delay of each part:

Two 100 Base-TX NICs = 100 bit times
Class II Repeater (Hub) = 92 bit times
Class II Repeater (Hub) = 92 bit times
5 Meter Cat-5 Cable Run = 5 x 1.112 = 5.56 bit times
80 Meter Cat-5 Cable Run = 80 x 1.112 = 88.96 bit times
100 Meter Cat-5 Cable Run = 100 x 1.112 = 111.2 bit times

Step 4: Add all delay values together

100 + 92 + 92 + 5.56 + 88.96 + 111.2 = 489.72 bit times

Step 5: Subtract result from 512

512 - 489.72 = 22.28

Result

Since the result is a positive number this network is valid and should work with no problems. Note that this network can't be expanded any farther, as just one more Class II hub will add more propagation delay than we can handle.

Larger-Scale Fast Ethernet Networks

The previous sections may lead one to believe that Fast Ethernet networks are not able to be scaled to support a large number of users or cover long distances. In a network built around conventional hubs this is an accurate assessment. However, there are techniques which can be employed to increase the scalability of the network considerably.

The first, and easiest, solution is to use only one repeater with the number of ports which need to be supported. Generally, this is accomplished by using stackable hubs. Stackable hubs are units which are interconnected with special "stack" ports and cables. All hubs in the same stack become part of one logical repeater, and the stack is counted as only one Class I or Class II repeater regardless of how many hubs are used to build it. For example, a stack consisting of ten hubs, each with twelve ports, behaves just like a single 120 port hub. Stackable hubs are an excellent solution for applications where all devices are within 330 feet of a centralized wiring location.

The second solution is to use Fast Ethernet switches. Basically, a switch is a multi-port bridge. Each port on the switch is in its own collision domain, and each subnetwork is calculated separately according to the rules above.

Sometimes a Fast Ethernet requires a long distance to be covered. An example would be interconnecting two buildings in a campus environment. The solution is to use Full Duplex technology and fiber optic cable. At each end, a Fast Ethernet switch is installed, and a fiber optic link is run from switch to switch. Since this link is run only from one point to another, and we have distinct circuits for transmit and recieve, there is no risk of collision occuring. Therefore we basically "turn off" CSMA/CD and let the switches transmit to each other at will. Propagation delay is not a factor in Full Duplex links.

It is important to note that Full Duplex can not be used in a shared-media environment, such as is created by a hub where there are more than two devices present on one collision domain. Full Duplex can only be run from switch to switch, DTE to DTE, or DTE to switch.

Conclusion

Fast Ethernet networks provide very high throughput, however they need to be carefully planned to ensure that they will work. The network can have no more than 512 bit times of round-trip delay between the two most distant nodes, and having more delay will bring the network to it's knees. If a network configuration fails the 512 bit time test, then the network needs to be segmented with local bridges or Fast Ethernet switches. Each segment must then be calculated independently to ensure that each has no more than 512 bit times of worst case propagation delay. Stackable hubs, switches, and full-duplex fiber optic links can provide a way to design an overall network which is larger than the 512 bit-time rule.