A baseline is used as a basis against which things are measured. Baselines have been traditionally used when you want to determine the effect of a change. For example, if you want to optimize your wireless network, you need to take a baseline of metrics such as retry rates or average data rates so that you can measure if the changes had a positive or negative impact.

A threshold is a level that must be exceeded to trigger an action. Thresholds are commonly used in network monitoring systems for alerts. For example, if a retry rate threshold were set at 50%, the system would trigger a warning when the retry rate exceeded 50%.

CV-CUE studies the behavior from the historical data of clients, APs and applications, automatically calculates a baseline. The baseline is calculated at an interval of 15 minutes. Any behaviour that deviates significantly from the baseline is considered to be an anomaly and highlighted in the graph. In controller based network monitoring systems, thresholds are static and the same value gets applied globally. This creates problems for network admins because wireless network characteristics can be different in different environments.

Thresholds are good for monitoring information where there is a clear, non-arbitrary delineation between acceptable and not acceptable. Thresholds are static. They do not adjust to changing conditions. Wireless networks are dynamic and change over time. The normal level of retry rates may be very different today and a month from now. Clients change, environments change, applications change, and usage changes rapidly. A static threshold is a challenge because it does not adapt to what is normal for the network. Then, if some metric regularly crosses its static threshold, the network admin is bombarded with irrelevant warnings. The network admin must then go in and reset the threshold. The problem lies in determining what the correct threshold is. If the threshold is set low, there will be too many alarms as to cause alarm fatigue. This is dangerous because valid alarms are lost in the sea of unimportant, false positive alarms. To counter alarm fatigue, many network admins set the threshold too high. This is dangerous because valid problems (false negatives) do not trigger action.

CV-CUE takes the idea of the baseline and makes it dynamic. Dynamic baselines determine what is normal for a network and adjust as network conditions change. For example, retry rates may be low when the Wi-Fi is first set up with only a few clients. Later, when many more clients are added to the Wi-Fi network, the retry rate may be very different. Dynamic baselines adjust as networks change. This avoids the problem of thresholds while allowing comparisons to the baseline to identify real problems.

The Baseline Graph has a provision to filter data. You can zoom in and zoom out the graph to view the granularity in detail. The zoom feature is at the bottom of the graph.

CV-CUE includes baselines for both connectivity and performance events. The table below lists the available baselines and where they can be found on the CV-CUE interface.

Example 2: Baseline - Data Rate

The following image displays the baseline graph for Data Rate:

The graph displays the calculated baseline of the average data rate consumed by an individual client. The anomalies are calculated by comparing the data rate against the globally configurable threshold. Data Rate is a metric where what is acceptable is not unique per network or environment so the use of a threshold to detect anomalies is appropriate. The baseline and deviation band are still calculated, but anomalies are determined by the data rate threshold.

Client connection success and failure with root cause analysis are reported by the AP to Arista Cloud almost immediately after it occurs. Performance and other data are aggregated and reported every 15 minutes.

The below table contains the attributes specifying the detailed info about the connected clients. The info is available in the tabular format on data point drill down from any baseline chart. The attributes with no specific name of a baseline chart are common for all the charts.