Previous Table of Contents Next


Hit Estimation

Perhaps the most difficult estimate to make is the number of hits that are expected to occur during the busiest hour of the business day. Access to an organization’s Web server depends on a large number of variables, many of which are beyond the control of the organization.

For example, although a company can control advertising of its Web’s URL in trade publications, it may be difficult (if not impossible) to inhibit robot search engines from visiting the site, retrieving each page available for public access on the company’s server and indexing the contents of the server’s Web pages. Once this index is placed onto the database of a search engine, access to the company’s Web server can result from persons invoking a Web search using Lycos, Alta Vista, or a similar search engine.

Unfortunately, because of limitations associated with many search engines, forward references to an organization’s Web server may not be relevant and can alter the distribution of page hits as many persons, upon viewing your home page, may click on the “back” button to return to the list of search matches provided by a search engine query and select a different match. If the organization was a tire distributor named Roosevelt Tires, for example, many Web search engines would return the home page URL in response to a search for the word “Roosevelt,” even though the person was searching for references to one of the presidents and not for an automobile tire distributor.

Many IAP can furnish statistics that may be applicable for use by an organization. A major exception to using such average statistics is if an organization is placing highly desirable information on the Web server, such as the results of major sports events like the Super Bowl or World Series as they occur. Otherwise, the information concerning busy hour hits the company’s IAP supplies can be considered to represent a reasonable level of activity that will materialize.

Returning to the estimation process, assume the organization can expect 660 hits during the busy hour. Although this hit activity may appear to be low in comparison to the tens of hundreds of thousands of hits reported by well-known URL representing popular Web server sites, during a 24-hour period you are configuring the operating rate of the WAN connection to support 24 X 660 or 15,840 hits, based upon a busy hour hit rate of 660.

According to statistics published by several IAP, during 1995 the average number of hits per Web site when the top 100 sites are excluded is under 5,000 per day. Thus, if an organization is the typical business, college, or government agency, it may be able to use a lower WAN operating rate than determined by this example.

After determining the number of hits the Web site will support during the busy hour and the average number of bytes that will be transmitted in response to a hit, it is possible to compute other WAN operating rates. For this example, each hit results in the transmission of 90,500 bytes, and the WAN operating rate is sized to support 660 hits during the busy hour. Thus, the results obtained are:

660 hits per hour × 90,500 bytes per hit × 8 bits60 minutes per hour
× 60 seconds per minute = 132,733 bps

LAN BANDWIDTH CONSTRAINTS

Based on the preceding computations it would be tempting to order a 192K-bps Fractional T1 as the WAN connection to the IAP, because the next lower fraction of service, 128K bps, would not provide a sufficient operating rate to accommodate the computed busy hour transmission requirement for 132,733 bps. However, before ordering the fractional T1 line, the business needs to consider the average bandwidth the Web server can obtain on the LAN it is connected to.

If the average bandwidth exceeds the computed WAN operating rate, the LAN will not be a bottleneck that should be modified. If the average LAN bandwidth obtainable by the Web server is less than the computed WAN operating rate, the local area network will function as a bottleneck, impeding access via the WAN to the Web server. This means that regardless of any increase in the operating rate of the wide area network connection, users’ ability to access the organization’s Web server will be restricted by local traffic on the LAN.

If this situation should occur, possible solutions are segmenting the LAN, creating a separate LAN for the Web server, migrating to a higher-speed technology, or performing a network adjustment to remove the effect of a portion of local LAN traffic functioning as a bottleneck to the Web server.

Determining the Effect on Local Traffic

To illustrate the computations involved in analyzing the effect of local traffic, assume the LAN shown in Exhibit 1 is a 10M-bps 10Base-T network that supports 23 workstations and one file server in addition to the Web server, resulting in a total of 25 stations on the network. This means that on the average, each network device will obtain access to 1/25 of the bandwidth of the LAN, or 400,000 bps (10M bps/25).

However, the bandwidth of the LAN does not represent the actual data transfer a network station can obtain. This is because the access protocol of the network will limit the achievable bandwidth to a percentage of the statistical average.

For example, on an Ethernet LAN that uses the CSMA/CD protocol, collisions will occur when two stations listen to the network and, noting an absence of transmission, attempt to transmit a frame at or near the same time. When a collision occurs, a jam signal is transmitted by the first station that detects the high voltage resulting from the collision, causing each station with data to transmit to invoke a random exponential backoff algorithm.


Previous Table of Contents Next

Copyright © CRC Press LLC