Analysis Topology Widget

Overview

The Analysis Topology Widget uses the Topology metadata on Sessions to graph connections between topology nodes.

Assumptions

Topology is assumed to be in the same direction for each Session object, for example, nodes are consistently provisioned in the metadata in order from the network core to edge. Any circular references detected between nodes are removed when building the graph (see the browser console logs for any cyclical connections removed).

Metrics

The graph is presented with a single active metric, selectable on the widget.

Nodes that appear in multiple Sessions are consolidated into a dataset for the node. The average value for the node dataset is used as the aggregate value of the node; the min/average/max along with the volume of rows in the node dataset can be seen when hovering a node. For count-based metrics, a sum aggregate is used instead of an average.

The size of the node is scaled relative to the aggregate value (rather than Session volume) to emphasize nodes with large metric values.

The widget can be navigated using the graph, but for more detailed/specific navigation, a table is provided with nodes ranked based on the nodes weighted diff from their parent nodes.

A weighted average is computed by taking the aggregate value (average/sum) of each parent node and scaling the aggregate according to the volume of Sessions present in the parent node. For instance, in the following graph, fredericton has 3 parent nodes: rtr1810, rtr1998 and rtr1809. rtr1810 and rtr1998 each have two sessions, while rtr1809 only has one session, so the weighted average would be:

average (253.827) * 0.4 +
average (104.990) * 0.4 +
average (29.797) * 0.2
= 149.4862

The weighted diff is the difference between the node’s average and this weighted average from the parent nodes.

fredericton average (149.486) -

weighted parent node average (149.4862)

= -0.002

This value isn’t seen as a significant difference compared to the diff between rtr1810 and its parent rtr0501, which is a much higher 104.341.

Colors/quantiles/bins

Because the weighted diff is a relative value, it’s hard to set firm boundaries for coloring. Instead, the widget uses quantiles, currently set at p50, p80 and p98, to partition the metrics between the min/max present in the dataset currently being presented. This means that if filters change, for example, selecting a node from the table, the quantiles will re-balance and colors may change.

Node aggregate values can use more firm value boundaries for coloring

Note that these color rules don’t have an impact on the diff colors, but will override the default quantile bins for node aggregate values. For example, in this screenshot, rtr1998 is now colored red instead of green when using the default quantile bins as seen in the prior screenshot. The diff colors remain unchanged.

Node interactions

Nodes in the chart can be hovered to show the breakdown of Sessions/metric ranges for the node.

Nodes within the table can be hovered to filter to the topology involving that specific node. Clicking to select the node will pin the node as a temporary filter, allowing for a more in-depth exploration of the filtered topology. The temporary filter can then either be dismissed, pushed to the Analysis as a global filter (Update Data will be required to rebuild the dataset), or used to create a reference line in the Timeseries Widget.