8.1. Charts Panel

In this panel will be displayed the different chartbox with the different series the user creates.

8.1.1. Chart series

A DataPoint in a chart series in the Fast DDS Monitor refers to a specific data type of the DDS network monitoring. Each DataPoint consists in a timestamp and a real value. The timestamp is the moment the data has been created, reported, received, etc. depending on the DataKind. The value refers to the real value of this DataPoint for this DataKind at the moment of the timestamp value.

For example, each DataPoint of the DataKind DATA_COUNT has a timestamp and an integer value. This value is the number of Data packets a DataWriter sent since the last time that same data was reported.

8.1.2. Chart view

Every DataKind is represented similarly inside a Chartbox. None, one or multiple lines of different colors will represent the different data series that are being displayed in the chart with the following format:

  • The X axis is the time value of the data.

  • The Y axis is the value that is store in the data.

This could be integers, doubles or times, but for the same DataKind will always be the same value type. Every point represents an accumulated value of DataPoints of a data kind in a time range.

8.1.3. Chartbox kinds

There are two kinds of Chartbox that could be created, historic and dynamic.

  • The historic chartbox represents static data that has been accumulated along the monitor execution. This data is not updated with new data or refreshed in case somethings has changed. See the details of this chartbox kind in section Historic Data.

  • The dynamic or real-time chartbox represents the new data that is being collected in real time by the monitor. These series will update over time with the new data received from a DDS network, and so the new data will be displayed in pseudo-real-time. See the details of this chartbox kind in section Real-Time Data.

8.1.4. Common Series parameters

Some of the parameters of the chartbox creations remains on the kind of chartbox to create. The ones mentioned here are the common parameters used by the two kinds of chartbox.

8.1.4.1. Data kind

The DataKind refers to the data that this chartbox will refer to. There are several DataKinds that represent each of the data kinds that a Fast DDS network can report. (Be aware that by default a DDS network will not report most of this data, and in case of Fast DDS it must be configured beforehand in order to report it periodically).

8.1.4.2. Series label

Name the new series in the Chartbox. If not set, the default series name is formed according to the following rule: <cumulative_function>_<source_entity_kind>-<target_entity_id>_<target_entity_kind>-<target_entity_id>.

8.1.4.3. Source Entity Id

This is the name and entity Id of the entity from which the data will be collected in the format <name>:\<<id>\>. This field has an entity kind to encapsulate the ids of the entities with the same kind, and make it easier for the user to search for the id required.

Each DataKind is related with one entity kind, normally a DataWriter or a DataReader, which are the producers of that DataKind. However, every entity kind is available to choose the data to represent. When an entity without this type of data is selected, the monitor will look for the entities contained in the specified entity that do report this specific DataKind. For example, in case the user wants to represent the number of packets (DATA_COUNT) transmitted by a Host, the monitor will look for all the DataWriters contained in that Host and report the total number of packets.

Continuing with the DATA_COUNT DataKind example, the monitor will find every DataWriter in this Host by finding every User in this Host, then finding every Process in these Users, finding every DomainParticipant in these Processes and finding every DataWriter in these DomainParticipants. Hence, the data displayed will be all the data that all those DataWriters have stored, accumulated by the cumulative function. Please refer to Entities section for more information on the monitor entities connections.

It is recommended to check some examples (see Example of usage) in order to better understand this functionality.

8.1.4.4. Target Entity Id

Note

Not every DataKind has a target entity, so the dialog will only show this field when the DataKind selected requires it.

This is the entity Id of the entity to which the data refers. This field works similar to the Source Entity Id. Some DataKind has a target entity to which the data refers, and this target must be of an specific entity kind. Choosing an entity of a different kind than the one this data requires will be solved by using the same mechanism explained in Source Entity Id. That is, searching for the correct entity kind by following the connections between entities.

i.e. To show the DataKind FASTDDS_LATENCY, that is reported by a DataWriter, and targeted to a DataReader, from an entity Host_1 to an entity Host_2, both of kind Host, the data displayed would be collected following the steps described below:

  • Get all the DataWriters of Host_1 by searching for its Users, their Processes, their DomainParticipants and their respective DataWriters.

  • Get all the DataReaders of Host_2 by searching for its Users, their Processes, their DomainParticipants and their respective DataReaders.

  • Get all the data from the DataWriters found that refer to the DataReaders found inside the interval set.

  • Accumulate the data using the cumulative function.

It is recommended to check some examples (Example of usage) in order to better understand this functionality.

8.1.4.5. Statistics kind

The cumulative function is used to accumulate the data points that fulfilled the conditions set in this configuration. When there are several data points to show in a single time frame these data points are transformed into one by a cumulative function. This function is the one set in Statistics kind.

The available methods to accumulate some data points into one are:

  • NONE: returns the DataPoint with the lower time.

  • MEAN: calculate the mean value of all the data points.

  • STANDARD_DEVIATION: calculate the standard deviation of all the data points.

  • MAX: returns the DataPoint with the maximum value.

  • MIN: returns the DataPoint with the minimum value.

  • MEDIAN: calculate the median of all the data points.

  • COUNT: returns the number of DataPoint in this time frame.

  • SUM: calculate the sum of all the data points.

In case the Number of bins is 0 the Statistics kind is not used as the data is not going to be accumulated.