Chapter 3: Estimating Network Traffic
Although there are a large number of hardware and software products that can be used to obtain network utilization information, the use of those products obviously requires an existing network. If you are planning the installation of a new network, you cannot use a performance monitor to measure network utilization until the network is established. At that time you might notice a high level of utilization, requiring the network to be subdivided through the use of a local bridge or switching hub to provide a better level of performance to local area network (LAN) users. Thus, it is important to understand how you can estimate network traffic and use the results of that traffic estimation process. Doing so provides you with the ability to determine whether or not you should consider the subdivision of the network via the use of a local bridge, switching hub, or similar communications device to enhance the performance level of each network segment.
This chapter first focuses attention on developing a methodology to estimate network traffic on a local area network. We will develop a traffic estimation worksheet and use that worksheet to project the average and peak traffic that could be carried on a network. Next, we turn our attention to the typical average and peak utilization levels of Ethernet and Token Ring networks, discussing utilization levels that, when exceeded, may indicate a decision criterion for network subdivision. This will be followed by an example that will illustrate how the use of a local bridge can make performance on each part better than if a network was not subdivided.
3.1 The Network Traffic Estimation Process
Assuming you do not have access to monitoring equipment to analyze an existing LAN or that your network is in a planning stage, you can develop a reasonable estimate of traffic by considering the functions each network user performs . To facilitate the traffic estimation process, you can group a number of network users together into a "workstation class" category and perform your computations for a single workstation. Then you can multiply the results of that computation by the number of workstations grouped into the specific workstation class to obtain an estimate of network traffic for a similar group of network users. You can then repeat this process for each workstation class you defined and add the traffic estimate for all workstation classes to obtain an estimate of the average traffic that will be carried on the entire network.
3.1.1 The Traffic Estimation Worksheet
As mentioned at the beginning of this chapter, you can expedite the traffic estimation process through the use of a worksheet. Table 3.1 illustrates the general format of a traffic estimation worksheet you can consider to facilitate your network traffic estimation process. Prior to illustrating how you can use the worksheet, let us first review the meaning of each entry in the worksheet.
| Workstation Class | Number of Stations | ||||
|---|---|---|---|---|---|
| Activity Performed | Message Size in Bytes | Number of Frames /Message | Frame Size in Bytes | Frequency/ Hour | Resulting Bit Rate (bps) [*] |
| Subtotal bps = | |||||
| Total for all workstation class: | |||||
 | |||||
3.1.2 Workstation Class
The workstation class identifies a specific category of LAN user. Normally you will use the occupational category of a group of persons for that entry. Typically, but not always, that entry reflects their average usage of the LAN. Examples of some common workstation classes include secretary, engineer, accountant , payroll clerk, application programmer, etc.
3.1.3 Number of Stations
The number of stations entry on the worksheet permits you to indicate the number of workstations that fall into a specific workstation class. Thus, you would normally have one worksheet for each defined workstation class.
3.1.4 Column Entries
Turning our attention to the column entries, you will note six specific column headings. The first column provides you with the ability to define the major activities performed by a workstation in the given class. Examples of major workstation activities include loading programs and data files residing on the network server, saving data files to the server, transmitting and receiving electronic mail, and directing print jobs to a network printer. Although the worksheet illustrated in Table 3.1 contains nine rows, you can add additional rows if you wish to define more than nine specific activities for a particular workstation class.
The Message Size column references the average number of bytes in a particular event performed by the defined activity. For example, suppose the transmission of electronic messages is expected to average 1800 characters . Thus, the message size in bytes would be entered as 1800.
Although several articles and technical manuals read by the author indicate the direct use of the message size, doing so results in a degree of inaccuracy. Although the use of a traffic estimation worksheet as its name implies results in an estimate of network traffic, we can reduce the degree of inaccuracy by considering the fact that Ethernet, Fast Ethernet, Token Ring, and other types of LAN frames have a degree of overhead. That degree of overhead can range in scope from a few percent for large message sizes to 30 to 50 percent or more for small- sized messages. The failure of the authors of those articles and technical manuals to consider the fact that a message must be encapsulated into one or more LAN frames and that each frame contains a number of fields that wrap around the encapsulated information results in their suggested computations having a built-in error. Thus, you must consider the number of frames used to carry a message as well as the frame size. Here, the latter includes the overhead fields that contain the preamble (Ethernet) or starting delimiter (Token Ring), source and destination addresses, and other frame information. As noted in Chapter 2, the overhead associated with Ethernet frames can vary from a low of 26 bytes when 38 or more bytes of information are carried in the data field to as many as 63 bytes when that field only carries one character of information. Under Gigabit Ethernet, up to 448 carrier extension symbols can be added to ensure that the minimum length of a frame is 520 bytes, to include the Preamble and Start of Frame Delimiter fields. This means that frame overhead when transmission occurs on Gigabit Ethernet will exceed that on Ethernet and Fast Ethernet until at least 494 data characters are carried in a frame. In comparison, a Token Ring frame that is not routed between rings has a fixed overhead of 21 bytes.
The fifth column contained in the traffic estimation worksheet is used to place your estimate of the frequency per hour in which the row activity is performed. Even if this activity is only performed once per day, such as the remote load of a diskless workstation, you can express the frequency on an hourly basis. For example, if you expect an activity to occur once a day and project the workstation user to work an eight-hour day, then the frequency per hour becomes 1/8, or 0.125.
The last column in the traffic estimation worksheet contains the resulting bit rate computed for the specific activity entered on the row. As indicated by the footnote at the bottom of the worksheet, the bit rate in bits per second (bps) is determined by the following equation:
Once you compute the resulting bit rate for each activity, you can sum the results of those computations to obtain the bit rate for all activities for one workstation in the class of workstations with which you are working. To complete your computation for the workstation class, you would then multiply the number of workstations in the class by the summed bit rate, as indicated in the lower portion of Table 3.1. Next, you would complete the traffic estimation process by computing a total bit rate for each remaining workstation class and then sum the total for each workstation class to obtain a network traffic estimate.
3.1.5 Developing a Network Traffic Estimate
Because the best way to illustrate the network traffic estimation process is by example, let us do so. In doing so, let us assume you want to install a Fast Ethernet LAN that will operate at 100 Mbps. Suppose this network is intended to support 87 design engineers ; 30 technicians; 40 general support personnel in accounting, personnel, finance, sales, and marketing; 10 managers; and 10 secretaries. Thus, you would probably consider completing a series of five traffic estimation worksheets to cover each of the five general categories of users planned for the network. As an alternative, you might decide to complete separate traffic estimation worksheets for accounting, personnel, finance, sales, and marketing personnel, instead of grouping them into a general support personnel class of network user. If you do this, you would then be required to complete a total of nine traffic estimation worksheets instead of five.
3.1.6 Using the Traffic Estimation Worksheet
Table 3.2 illustrates the completion of a traffic estimation worksheet for the design engineer class of workstation users. In completing this worksheet, it was assumed that the average design engineer will perform a core set of seven LAN functions. Those functions are listed under the Activity Performed column in Table 3.2. The functions performed range in scope from loading a program that resides on a network server to loading and saving graphic images, sending and receiving messages, and printing both graphic images and text data on a printer to be connected to the Fast Ethernet network.
| Workstation Class Engineers | Number of Stations | 214 | |||
|---|---|---|---|---|---|
| Activity Performed | Message Size in Bytes | Number of Frames/Message | Frame Size in Bytes | Frequency/Hour | Resulting Bit Rate (bps) [*] |
| 1. Load graphic | 512 | 350 | 1526 | 16 | 2373 |
| 2. Save graphic | 1024 | 700 | 1526 | 8 | 2373 |
| 3. Send e-mail | 1 | 1 | 1526 | 4 | 1 |
| 4. Receive e-mail | 2 | 2 | 1526 | 4 | 3 |
| 5. Load program | 1024 | 700 | 1526 | 3 | 890 |
| Subtotal bps = | 5640 | ||||
| Total for all workstation class: | |||||
 | |||||
3.1.6.1 Required Computations
To illustrate the computations used for the completion of the traffic estimation worksheet, let us consider the load program activity entry row. Here it was assumed that programs which occupy an average of 640 Kbytes of storage on the network server will be loaded four times each hour. Although this program loading frequency may appear high, many engineering programs actually consist of a series of overlays and the loading of one program could easily result in the loading of several modules each hour as the design engineer accesses different program features in performing the design effort.
If the size of the program or program module loaded into the workstation is 640 Kbytes, then the actual size of the program is 640 * 1024 bytes per K, or 655,360 bytes.
On an Ethernet network, the maximum length of the information field in a frame that contains the program data transported from the server to the workstation is 1500 bytes. Thus, 655,360 bytes in the program divided by 1500 bytes that can be carried in a frame results in 437 frames that will be required to transport the program. Although the information field is 1500 bytes, the actual frame size will be 1526 bytes because there are 26 overhead bytes in an Ethernet frame. Based on the preceding , the resulting bit rate attributed to the load program activity becomes:
If you anticipate installing a Token Ring network, you would consider a data field size ranging from a minimum of 1 byte to a maximum of 4500 bytes for a 4-Mbps Token Ring network and to a maximum of 1800 bytes for a 16-Mbps network. Because the overhead of a Token Ring network is fixed at 21 bytes, you would compute the number of frames required to transport each type of message. Next, you would add 21 bytes to the frame size and perform a computation similar to the one just illustrated for the load program activity for the design engineer workstation class using a Fast Ethernet network.
In performing the previous computation, readers will note that it was not precise because the number of frames was rounded to 437 and did not consider the fact that the last frame does not actually carry 1500 bytes. For file transfers, program loads, and print jobs, you can safely disregard the fact that the last frame used to carry an activity may have an information field whose length is less than 1500 bytes. This is because doing so results in a maximum divergence of 4 bps over an hour for the activity being computed for each occurrence of the activity. For example, 1500 bytes * 8 bits/byte results in an Ethernet information field carrying a maximum of 12,000 bits, which when divided by 3600 seconds per hour is less than 4 bps. Thus, your activity projection would be off by a maximum of 16 bps, which is far less than 1 percent. Because you are estimating network traffic, a good rule of thumb to follow is to ignore the fact that the last frame's information field used to carry an activity may be partially filled unless the activity is performed more than 50 times per hour. The latter may represent the use of an application performed by clerical personnel entering batches of data concerning personnel file updates, corporate receipts and disbursements, and similar types of work.
If you look at the entries in Table 3.2 and take the time to perform your own computations, you will note that the frame size for the activities "send message," "receive message," and "print text data" precisely represent the amount of data transmitted for each activity. This is because the message size resulted in only one frame being required to support the indicated activity, which results in a precise frame length being used.
3.1.7 Network Printing Considerations
One item worthy of mention concerning the activity entries in Table 3.2 involves network printing. If your network infrastructure is designed so that file servers function as print servers as shown in the top portion of Figure 3.1, the computations shown for network printing are correct. However, many networks are constructed using separate print servers. When this occurs, a print job transmitted by a network user first flows from his or her workstation to the file server, where it is placed in a print queue. Then, the file server transmits the print job from the queue to the print server as illustrated in the lower portion of Figure 3.1. Under this network printing configuration, the print job is transmitted twice, first from the workstation to the file server and then from the file server to the print server. If your network infrastructure will use separate print servers, you would then double the resulting bit rate for each print job computed in a traffic estimation worksheet.
Figure 3.1: The Method of Printing Can Double Network Traffic, Resulting from the Execution of a Print Job
To further illustrate the completion of the traffic estimation worksheet illustrated in Table 3.2, let us examine the load and save graphic image entries and the send and receive message activities. In addition, to further illustrate the computations involved in completing the worksheet, the resulting bit rate computations for each activity will also indicate the significant difference in the effect of graphic versus text information transfer on a LAN.
3.1.7.1 Graphic versus Text Transfer
Because the entries for loading and saving a graphic image are the same, you can compute one entry and use the resulting bit rate computation for each activity. Similarly, you can do the same for sending and receiving messages.
For each graphic image activity, let us assume the image consists of a file of 1024 Kbytes. Thus, the file size is 1024 * 1024 bytes per K, or 1,048,576 bytes. When carried by a 1500-byte Ethernet information field, this results in the use of 1,048,576 bytes/1500 bytes/frame, or 700 frames, to transport the graphic image. Because there are 26 overhead bytes in each frame, the actual frame length is 1526 bytes. Based on a frequency of 20 images loaded or saved per hour, each graphic image activity bit rate computation is as follows :
Now let us turn our attention to the send and receive message activities. For each activity, we assumed the message size is.5 K, or 512 bytes. Thus, the entire message can be transported by one Fast Ethernet frame that has a 1500-byte information field. Because there are 26 overhead bytes per Fast Ethernet frame, the frame size is 538 bytes. Based on a frequency of two messages sent or received per hour, each message activity's bit rate computation becomes:
In comparing the resulting bit rate of a graphic image activity to an electronic message activity, note the significant difference between the resulting bit rate of each activity. Here, each graphic image activity is in excess of 1500 times the resulting bit rate associated with a message. This comparison is performed by dividing the bit rate of 47,475 for 20 image operations per hour by 20 to obtain a per-image hourly bps rate of 2374. Performing a similar operation for a send or receive message results in a per-message hourly rate of 1.5. Then, the ratio of 2374 to 1.5 results in a value of 1582. Even if the message was doubled in size to 1024 bytes, the graphic image would have a bit rate approximately 800 times that of the message. If you consider the fact that a super VGA monitor color graphic image can easily exceed 1 Mbyte and that most electronic mail messages are relatively short, consisting of one or two paragraphs and typically less than 100 words, this explains why electronic mail has a negligible effect on LAN performance in comparison to the transmission of graphic images.
Returning to Table 3.2, note that the subtotal bps of 112,758 represents the average network traffic for one workstation used by a design engineer. Because there are 87 workstations used by design engineers, the total bit rate is 112,758 * 87, or 9,809,946 bps. Now that we have completed the traffic estimation worksheet for one workstation class, we would perform similar computations for each of the remaining four workstation classes. To facilitate our analysis, let us assume that the results of the computations for the five workstation classes are as summarized in Table 3.3.
| Workstation Class | Bit Rate (bps) |
|---|---|
| Design engineer | 9809946 |
| Technicians | 5422500 |
| General support personnel | 3238750 |
| Managers | 283920 |
| Secretaries | 1146750 |
| Total estimated bit rate | 19901866 |
3.1.8 Network Utilization Considerations
In examining the projected network traffic summarized in Table 3.3, note that the total bit rate of 19.9 Mbps represents a projected utilization level of 19.9/100.0, or 19.9 percent, of the available network bandwidth. If you feel a projected 19.9 percent level of network utilization is low, you are correct in your assumption for both Ethernet and Token Ring networks. However, as the level of network utilization increases , the effect on network performance depends on the type of network. A Token Ring network's transmission capability depends on several variables , to include the number of stations on the network, the average length of frames transmitted on the network, and the total length of cable used to form the network.
Under certain variable relationships, the capability of a Token Ring network to transport data can be as low as one half of the network's operating rate. Although Chapter 8 contains specific information concerning the development and execution of a mathematical model to determine the information-carrying capacity of a Token Ring network, we can generalize the findings of that chapter by noting that, when possible, you should attempt to keep the utilization level of a Token Ring network to a maximum of 70 percent of its operating rate. Doing so will preclude the occurrence of excessive response times. For an Ethernet LAN to include Fast Ethernet, the utilization level should be kept to 40 to 50 percent of the network's operating rate. The rationale for an Ethernet LAN having a lower utilization level threshold than a Token Ring network is due to the difference in the access method used by each network. Ethernet access is not predictable and can result in collisions that require a random time delay during which no station on the network can transmit data. In comparison, network access on a Token Ring network is predictable because a station can only transmit when it is able to acquire a free token.
A second difference between the transmission capability of each network concerns the method of frame transmission used on each network. On an Ethernet network there is a minimum fixed delay period of time between frame transmissions. In comparison, transmission can occur on a Token Ring network as soon as a station acquires a free token. Readers are referred to Chapter 7 for specific information concerning the flow of data on an Ethernet LAN.
3.1.9 Planning for Network Growth
In addition to estimating traffic that will be carried by a network, it is equally important to consider network growth. Doing so permits you to determine if a network structure will be able to accommodate a buildup in workstation usage of the network and/or an increase in network users over a period of time.
To illustrate how you can consider network growth, let us assume your organization anticipates additional hiring that will increase the design engineering staff by 20 percent, management personnel by 25 percent, and the secretarial staff by 5 percent. To consider the potential effect of adding workstations to the network to accommodate the additional hiring, you could use a network growth worksheet similar to the one contained in Table 3.4. This worksheet was completed based upon the projected network traffic contained in Table 3.3 and the previously discussed projections for additional employees .
| Workstation Class | Estimated Bit Rate (bps) | Projected Growth Rate (%) | Projected Bit Rate (bps) |
|---|---|---|---|
| Design engineer | 9809946 | 20 | 11771935 |
| Technicians | 5422500 |
| 5422500 |
| General support personnel | 3238750 |
| 3238750 |
| Managers | 283920 | 25 | 354900 |
| Secretaries | 1146750 | 5 | 1204087 |
| Total projected bit rate | 21992172 |
EAN: 2147483647
Pages: 111