The CAST Application Analytics Dashboard (AAD) is currently supported and tested for a use case of 200 applications maximum (see Installing and configuring the CAST Application Analytics Dashboard). For customers that have more than 200 applications to consolidate into the dashboard, this is page describes various good practices and configuration options that could help to reduce the risk of performance issues.
The main areas that can impact performance are as follows:
- Infrastructure (the hardware used to run the dashboard)
- Number of users that are going to connect to the dashboard and the scope of applications they are authorised to access
- Number of Applications that have been consolidated
- Number of snapshots
- Number of tiles configured for display in the homepage
With regard to the deployment data sizing, please refer to Deployment - sizing to evaluate the appropriate configuration based on the number of Applications you will need to consolidated.
CAST Storage Service (CSS) installation
Install on Linux
To improve the performance of CSS, consider installing it on a Linux operating system instead of on Windows. See PostgreSQL for Linux or Docker for more information.
Isolate Measurement Service database
If the CAST Application Analytics Dashboard is going to be accessed by a high number of users, good practice is to isolate the Measurement Service database on a dedicated CAST Storage Service instance instead of installing it on the same instance used to host the Dashboard Service databases. This configuration provides improved performance with regard to the number of connections to be configured between the CAST Application Analytics Dashboard web application on Apache Tomcat and the CSS instance (see section below for more information about Tomcat configuration)
Run optimization procedures on CSS
It is good practice to run optimization procedures on the CAST Storage Service instance that is hosting your Measurement Service database on a regular basis or at least each time a Measurement Service database is deleted and restored in order to recover disk space used by deleted data. Please run the following commands:
Deployment on Linux
To improve performance, consider deploying Apache Tomcat on a Linux operating system instead of on Windows. Currently CAST does not provide any documentation around this, but Linux has been used successfully to host Apache Tomcat for use with the CAST Application Analytics Dashboard.
When deploying the CAST Application Analytics Dashboard consider modifying the default configuration in the %CATALINA_HOME%\webapps\CAST-AAD\META-INF\context.xml file to manage the number of connections between the web application (CAST AAD) and the CAST Storage Service instance. Below is the default configuration provided "out-of-the-box" in the context.xml file for both Tomcat 7 and 8/8.5:
Please optimize the following parameters depending on the number of users and user connection profile:
|Tomcat 7||Tomcat 8||Description|
(int) The initial number of connections that are created when the pool is started. Default value is 5.
|maxActive||maxTotal||(int) The maximum number of active connections that can be allocated from this pool at the same time. The default value is 20.|
(int) The maximum number of connections that should be kept in the pool at all times. Default value is 10. Idle connections are checked periodically (if enabled).
(int) The maximum number of milliseconds that the pool will wait (when there are no available connections) for a connection to be returned before throwing an exception. Default value is -1 (i.e. the pool will wait indefinitely).
The configuration of these parameters is easier to understand if you think of them as exactly corresponding to the number of cash registers you need to open to avoid having too many customers waiting in your shop to buy articles. Depending on the number of customers and period in the day, you will need to adapt the number of opened cash registers. In addition to that, these cash registers can be used for other shops in parallel. It is therefore very important to know the users' profile and when they are going to connect to your dashboard to be able to adapt these parameters:
- Considering AAD with a lot of potential users that can connect to a unique Measurement Service database, this default configuration should be good.
Using ping to check for performance issues
If you are facing performance issues, you can run a "ping" test to check the connection between the server hosting Apache Tomcat and the server hosting CSS/Postgres. This may help you determine whether your performance issues are due to a bottleneck on your internal network. Please proceed as follows:
- Open a command line window (cmd) from the server hosting Apache Tomcat
- Run the following command: ping <name_of_CSS_or_postgres_server>
If the response time returned during the ping test is high, the bottleneck/issue is likely to be located on the internal network. Please contact your network administrator in this case.
Applications and Snapshots
The number of Applications consolidated into one single Measurement Service schema can impact the performance of your dashboard. CAST generally considers more than 200 Applications to be a "large" deployment where performance issues may start to appear, but this figure is not set in stone and depending on other criteria (such as the number of snapshots per Application) you may see a performance hit with a lower number of Applications (for example 20 Applications with a total of 2000 snapshots).
The number of snapshots per Application is also key as mentioned already. You may only have 20 Applications, but if the total number of snapshots in these 20 Applications is 2000 (for example) then you may see a performance hit.
The number of users that are connecting to your dashboard can impact performance:
- If you are using Active Directory/LDAP authentication, performance of the dashboard can be adversely affected by the response time of your Active Directory/LDAP server.
- If all users are authorized to view all Applications, performance can also be impacted
Checking Active Directory/LDAP authentication response time
If you are using Active Directory/LDAP authentication, performance of the dashboard can be adversely affected by the response time of your Active Directory/LDAP server. To check whether this is the case, please proceed as follows:
- If you already have a HAR (HTTP Archive) trace file, you can use the HAR files viewer with Google Chrome: http://ericduran.github.io/chromeHAR/
- If you do not have a HAR (HTTP Archive) trace file, you can view response times by:
- connecting to the CAST Application Analytics Dashboard using Google Chrome, tapping F12 and then reproducing the issue. You will get a HAR trace at the bottom of the page in the "Network" tab.
- In the HAR trace look for the following as shown in the images below:
- The value listed in "ms" in the "Time" column indicates how login the process has taken. If the value is high, please contact your network administrator in this case as this delay is being caused by something other than the CAST dashboard. For example: the total login time to the dashboard is less than 6 seconds but you observe that the "rest/ login" time is equal to 5 seconds. In this case, the issue is located on your Active Directory/LDAP server.
When you are consolidating a large number of Applications, you should always actively use the data authorization feature to prevent users from viewing Applications/data they do not need to, therefore improving performance.
- When a user is logged in, if authorization mapping between user and Application has been implemented, the user will access only the Applications he/she is authorized to access. As such the number of Applications that are accessed will be highly reduced and performance will improve (less Applications to display, less calculations to provide in the homepage, less data to load in graphs)
- Using data authorization via tags is one way to avoid having to duplicated CAST Application Analytics Dashboard instances (database + war).
Each tile that is configured for display in the dashboard home page will take time to render as calculations will have to be done based on the number of applications and snapshots (e.g. to display TQI score evolution tile, the average of all snapshots to all applications must first be calculated and then the data must be displayed graphically).
Reduce the number of tiles displayed in the homepage (cmp.json)
Some tiles can be time consuming to load. Please use the following as guide to limit the number tiles in your homepage:
- Use a maximum of one tile for benchmarking (eg bubble graph tile or treemap tile). Avoid using this type of tile if you can as it is time consuming to load.
- Don't use a tile that lists rules across all Applications
- Remove or reduce the number of evolution tiles
Provide a direct filtered URL to reduce number of Applications that load in the home page
You can provide a direct URL that points to a filter (i.e. a tag or a specific time filter). This will improve performance as not all the Applications in the portfolio need to be displayed:
To share a filtered URL simply browse to the filtered data you want to share and copy the URL from your browser's address bar, e.g.:
Performance test results
Some performance tests have been completed based on specific scenarios for the CAST Application Analytics Dashboard. These tests provide some insight into the limit of responses time depending on the number of simultaneously connected users.
- See Last Performance analysis report with JMeter on the CAST Application Analytics Dashboard for CAST AIP 7.3: AAD Performance Analysis (June 2014).pptx
- See Performance Analysis report with JMeter on the CAST Application Analytics Dashboard for two different Measurement Service databases: AAD Performance Analysis (May 2016).pptx