When LANs had only a few users, performance was usually very good. Today, however, when most computers in an organization are on LANs, performance can be a problem. Performance is usually expressed in terms of throughput (the total amount of user data transmitted in a given time period). In this section, we discuss how to improve throughput. We focus on dedicated-server networks because they are the most commonly used type of LANs, but many of these concepts also apply to peer-to-peer networks.
To improve performance, you must locate the bottleneck, the part of the network that is restricting the data flow. Generally speaking, the bottleneck will lie in one of two places. The first is the network server. In this case, the client computers have no difficulty sending requests to the network server, but the server lacks sufficient capacity to process all the requests it receives in a timely manner. The second location is the network circuit, connecting the LAN to the corporate BN. In this case, the server can easily process all the client requests it receives, but the circuit lacks enough capacity to transmit all the requests to the server. It is also possible that the bottleneck could lie in the client computers themselves (e.g., they are receiving data too fast for them to process it), but this is extremely unlikely—unless, of course, you are still using old computers!
The first step in improving performance, therefore, is to identify whether the bottleneck lies in the circuit or the server. To do so, you simply watch the utilization of the server during periods of poor performance. If the server utilization is high (e.g., 60 to 100 percent), then the bottleneck is the server; it cannot process all the requests it receives in a timely manner. If the server utilization is low during periods of poor performance (e.g., 10 to 40 percent), then the problem lies with the network circuit; the circuit cannot transmit requests to the server as quickly as necessary. Things become more difficult if utilization is in the midrange (e.g., 40 to 60 percent). This suggests that the bottleneck may shift between the server and the circuit depending on the type of request, and it suggests that both should be upgraded to provide the best performance.
Leave a Reply
You must be logged in to post a comment.