Figure 1. Bandwidth Utilization Graphs do not depict network performance. Five minute averages are too long to capture temporary congestion.
Figure 2. Typical router timeslot usage over 1 second.
This graph shows how a router always transmits at its full interface speed but varies its send time in order to throttle bandwidth. As long as there are free timeslots to send traffic, there is no congestion. If all 40 timeslots were filled with green lines then that link would be said to be saturated.
Figure 3. A typical Network Utilization & Latency graph shows increased latency with increased traffic. There should be a relationship between latency and traffic. If latency is increasing without the corresponding volumes of traffic then further investigation may be required.
MONITORING NETWORK PERFORMANCE
There are many misconceptions about monitoring network performance. One common misconception is that you can look at a bandwidth utilization graphs and determine the quality of the network performance. This is not the case, busy links don’t necessarily perform poorly and quiet links are not necessarily fast. Bandwidth utilization graphs are poor indicators of network performance for the following reasons:
Utilization graphs are usually plotted based on 5 minute averages. For a router, even a single second is a very long time, in fact a router usually divides each second into approximately 40 time slots depending on the interface type/speed (see figure 2). A 5 minute average is too long of a time period to assess whether any congestion was experienced because congestion is really the measure of consecutive router timeslots receiving more traffic than they are capable to send.
Delays and packet drops occur when the router has insufficient time slots to service all of the outgoing traffic. The router will begin to queue excess traffic, but once the queue is full excess traffic is then dropped. Very bursty traffic (traffic that arrives all at once in large quantities then stops suddenly) can cause delays and drops while avoiding detection on a monitoring system because the monitoring system averages over a long period of time do not reflect the temporary congestion.
Dropped Packets and Their Effect of Network Monitoring
One obvious drawback from dropped packets is the need to retransmit those packets. But TCP/IP also has some flow control mechanisms that cause additional delays. These delays include:
1. Delay the retransmission using a timer
2. Send smaller packets in order to decrease the rate of transmission.
What we are trying to say is that it is feasible that an inconspicuous low utilization graph is masking what may have been quite a slow tedious response for an end user. Because the link is dropping packets, causing the sender to throttle its delivery, the throughput is genuinely low but the cause of the low throughput is not captured in the utilization graphs.
Figure 4. The timeslot graph shows actual router utilization during a short period of time compared to the network utilization graph above. Although the network graph showed 50% utilization during the period, the actual router itself is seen to be almost 100% utilized over the first 12 timeslots and almost 0% utilized the remainder of the time.The user experience would have been severely affected for computers requesting bandwidth during the busy period even though the utilization graph suggests that there was capacity available.
Custom Search
