Signs that an Ethernet LAN Upgrade Is Due
Until recently, LAN congestion was rarely a problem. LANs could move more data per second than computers
could output. Now, advances in computing and commu-nications technology have shifted the bottleneck from
the computer to the LAN. The output from one device can occupy the full channel capacity of a typical LAN.
When several such devices share the channel, congestion often results.
To determine if your LAN is congested, consider the following factors.
Higher Network Utilization
Measure network utilization, sometimes called channel utilization, which is the percentage of time that a commu-nications path is busy carrying data. The higher the traffic load, the higher the network utilization. You can measure network utilization by using network monitoring equip-ment, such as protocol analyzers or Remote Monitoring
(RMON) devices. Certain hubs and switches provide net-work utilization statistics on their visual displays. For many common environments, including office automa-tion LANs with dozens of stations, the LAN is approach-ing an excessive load when network utilization reaches the following levels:
- 20 percent averaged over an eight-hour work day
- 30 percent averaged over the worst hour of the day
- 50 percent averaged over the worst 15 minutes of the day
These figures are only guidelines. Some application envi-ronments may operate well under heavier loads or fail at
lighter levels. Network performance will still vary based upon the number of stations on the LAN, application
behavior, and traffic patterns.
Note that for very short-term periods (seconds, or even tens of seconds), network utilization may be nearly 100
percent without causing any problems. Temporary high utilization might occur during a large file transfer between
a pair of high-performance stations on an otherwise quiet network.
Increased Collision Counts
If multiple stations on a shared network begin sending Ethernet packets at the same time because they all sense a
quiet network, a “collision” between packets results. When a collision occurs, each participant waits a random
amount of time and tries to send its packet again. An increase in the number of packet collisions on an Ethernet
LAN can indicate rising LAN congestion. Collisions do not always indicate a congestion problem, however. The
Ethernet protocol uses collisions to quickly redistribute the traffic load over the available time, maximizing channel
utilization and application throughput. In fact, collisions consume a very small percentage of available channel
capacity, even under a moderate to heavy traffic load. If user performance and application throughput are accept-able,
you can pay little attention to collision statistics, and regard them as only a minor factor in network health.
Application Performance Degradation
When the LAN is congested, applications do not perform as well as they do otherwise. File transfers take longer, and
terminal sessions are sluggish. In extreme circumstances, sessions may time out and disconnect and applications or
operating systems may actually crash, requiring a system restart. A decrease in application performance can be an
indication of LAN congestion. However, it can also result from other factors, such as CPU, memory, disk perfor-mance,
and number of users.
Increased Network Delay
All LANs have an upper limit on the amount of data they can carry. When temporarily overloaded, the LAN takes
longer to distribute the load. Therefore, when the load is heavy, the average delay, also known as service time,
increases. This delay makes the network seem slower to users. To directly measure service time, you need special
driver software. It is easier to assess congestion using other, more observable criteria, such as the data from network
analyzers.
User Dissatisfaction
The ultimate sign of LAN congestion is user dissatisfac-tion. If users are happy, then the network is functioning at
acceptable levels. Similarly, if users are dissatisfied with the LAN performance, this fact overrides all statistics indi-cating that performance is within acceptable ranges.
Note that user dissatisfaction with network performance does not always indicate a congestion problem. The
source of the problem might instead be applications, servers, protocol stacks, or WAN access devices—not just
the underlying LAN hardware and software.
