5.6 Applying the Equipment Sizing Process

Many methods are available for end users to obtain data traffic statistics required for sizing communications equipment. Two of the most commonly used methods are based on user surveys and computer accounting information.

End-user surveys normally require each user to estimate the number of originated calls to a network access point for average and peak traffic situations as well as the call duration in minutes or fractions of an hour , on a daily basis. By accumulating the traffic data for a group of users in a particular geographic area, you then can obtain the traffic that the access controller will be required to support.

Suppose a new application is under consideration at a geographic area currently not served by a firm's data communications network. For this application, ten PCs with the anticipated data traffic denoted in Table 5.17 are to be installed at five small offices in the greater metropolitan area of a city. If each PC user will dial a centrally located LAN access controller, how many dial-in lines, auto-answer modems, and access controller ports are required to provide users with a 98 percent probability of accessing the network upon dialing the LAN access controller? What would happen if a 90 percent probability of access were acceptable?

Table 5.17: PC Traffic Survey
	Calls Originated per Day		Call Duration (minutes)
PC	Average	Peak	Average	Peak
A	3	6	15	30
B	2	3	30	60
C	5	5	10	15
D	2	3	15	15
E	2	4	15	30
F	2	4	15	30
G	3	3	15	35
H	4	6	30	30
I	2	3	20	25
J	2	2	15	60

For the ten PCs listed in Table 5.17, the average daily and peak daily traffic are easily computed. These figures can be obtained by multiplying the number of calls originated each day by the call duration and summing the values for the appropriate average and peak periods. Doing so, you obtain 480 minutes of average daily traffic and 1200 minutes of peak traffic. Dividing those numbers by 60 results in 8 erlangs average daily traffic and 20 erlangs peak daily traffic.

Prior to sizing, some additional knowledge and assumptions concerning PC traffic will be necessary. First, from the data contained in most survey forms, information regarding busy-hour traffic is nonexistent, although such information is critical for equipment sizing. Although survey forms can be tailored to obtain the number of calls and call duration by specific time intervals, for most users the completion of such precise estimates is a guess at best.

Busy-hour traffic can normally be estimated accurately from historical or computer billing and accounting type data, or from the use of a network management system that logs usage data. Suppose that the use of one of those sources shows a busy-hour traffic equal to twice the average daily traffic based upon an eight-hour normal operational shift. Then the traffic would be (8/8) * 2 or 2 erlangs, while the busy-hour peak traffic would be (20/8) * 2, or 5 erlangs.

The next process in the sizing procedure is to determine the appropriate sizing formula to apply to the problem. If we assume that users encountering a busy signal will tend to redial the telephone numbers associated with the access controller, the Poisson formula will be applicable . From Table 5.18, the 7.0- erlang traffic column shows a 0.01656 probability (1.65 percent) of all channels busy for a device containing six channels, 0.05265 for five channels, and 0.14288 for four channels. Thus, to obtain a 98 percent probability of access based on the daily average traffic would require six channels, while a 90 percent probability of access would require five channels.

Table 5.18: Poisson Distribution Program Extract
	Traffic in erlangs
Port No.	5.50	6.00	6.50	7.00	7.50
1	0.39347	0.63212	0.77687	0.86466	0.91791
2	0.09020	0.26424	0.44217	0.59399	0.71270
3	0.01439	0.08030	0.19115	0.32332	0.45619
4	0.00175	0.01899	0.06564	0.14288	0.24242
5	0.00017	0.00366	0.01857	0.05265	0.10882
6	0.00001	0.00059	0.00446	0.01656	0.04202
7	0.00000	0.00008	0.00093	0.00453	0.01419
8	0.00000	0.00001	0.00017	0.00110	0.00425
9	0.00000	0.00000	0.00003	0.00024	0.00114
10	0.00000	0.00000	0.00000	0.00005	0.00028
11	0.00000	0.00000	0.00000	0.00001	0.00006
12	0.00000	0.00000	0.00000	0.00000	0.00001
Probability that all ports are busy when call attempted, which is the grade of service.

If we want to size the equipment based on the daily peak traffic load, how would sizing differ ? We now would use a 5-erlang traffic column contained in the sizing tables. From the table, 11 channels would provide a 0.01369 probability (1.37 percent) of encountering a busy signal, while 10 channels would provide a 0.03182 probability. To obtain a 98 percent probability of access statistically would require 11 channels. Because there are only ten terminals, logic would override statistics and ten channels, or one channel per personal computer, would suffice. It should be noted that the statistical approach is based on a level of traffic that can be generated from an infinite number of computers. Thus, you must also use logic and recognize the limits of the statistical approach when sizing equipment. Because a 0.06809 probability of encountering a busy signal is associated with nine channels and a 0.13337 probability with eight channels, nine channels would be required to obtain a 90 percent probability of access.

In Table 5.19, the sizing required for average and peak daily traffic is listed for both 90 percent and 98 percent probability of obtaining access. Note that the difference between supporting the average and peak traffic loads is four channels for both the 90 percent and 98 percent probability of access scenarios, although peak traffic is 2.5 times average traffic.

Table 5.19: Channel Requirements Summary
Probability of Access (%)	Daily Average	Traffic Peak
90	5	9
98	6	10

The last process in the sizing procedure is to determine the number of channels and associated equipment to install. Whether to support the average or peak load will depend on the critical nature of the application, funds availability, how often peak daily traffic can be expected, and perhaps organizational politics. If peak traffic occurs once per month, we could normally size equipment for the average daily traffic expected. If peak traffic was expected to occur twice each day, we would normally size equipment based upon peak traffic. Traffic between these extremes may require that the final step in the sizing procedure be one of human judgment, incorporating knowledge of economics, and the application into the decision process.