Install analytics module for SAYMON ISO image
By default, the analytics module used to determine splashes and predictions is not included in the SAYMON ISO image. To install the analytics module:
Download analytics-create-and-run.sh script and place it in the file system on the virtual machine on which the SAYMON ISO image is installed.
Run the script:
x.y.z - analytics module version compatible with your SAYMON version ;
To check whether the "saymon-analytics" container is up and running, use the command
Connect analytics module to SAYMON in configuration files
In order to connect the analytics module to SAYMON server, the SAYMON configuration files must be modified:
To the section Server of the server configuration file add the following parameters:
In the client configuration file set the parameter to :
Restart the SAYMON server:
Analytics module monitoring
Logs of the saymon-analytics container can be used in order to monitor work of the analytics module. The steps are:
Connect to the container with the command:
Open the log file:
Log file contains information about
- the settings with which the module is started (the socket and the number of processes involved in the pool),
- the processed metrics,
- the results of processing metrics,
- errors that occur during operation of the the module.
Analytics module performance
Analytics for each object is carried out in a separate stream of calculations using one logical core of the system. All values of metrics of the same object are processed sequentially, which allows sequential update of the forecast model in accordance with the order of data receipt. As a rule, the performance of modern processors allows to process each incoming value in real time with the maximum frequency is equal to one metric value per second. However, on superheated or overloaded machines the performance of the cores can decrease, it leads to accumulation of tasks for the analytics module and increasing the reaction time of the system to new incoming data. To increase system performance in tasks with high frequency of incoming metrics, it is recommended to allocate one logical core for each object with analytics. If there are additional cores, the system can also use them for preprocessing analytic tasks, since locking on a predictive model takes only a fraction of the total processing time of an analytic task.