Appboard/2.6/admin/performance and sizing

Revision as of 10:59, 30 April 2015 by imported>Jason.nicholls (1 revision)


AppBoard performance is heavily influenced by the size of data-sets, caching times, client polling frequencies, and the miscellaneous ways to transform data such as sub-queries, filtering, and data processing scripts. In addition, if using AppBoard in conjunction with enPortal proxied content then this will also introduce additional load.

Refer to the System Requirements page for suggested server sizing depending on the deployment scenario. Typically the suggested sizing should handle hundreds of simultaneous users on the system - but this does vary and additional monitoring and tuning may be required for your particular situation.


Template-tip.png
Setting up system and application level memory and CPU monitoring is a recommended best practice to keep an eye on AppBoard performance.

Memory

As with any Java application performance is greatly impacted if there is insufficient memory and greater amounts of CPU time is given up trying to free memory until the point it becomes too slow and OutOfMemoryErrors occur.

While it's possible to profile, analyse and determine the ideal amount of memory to allocate this can be a lengthy process.

As a general rule, if experiencing performance issues and/or out of memory errors then increase the memory allocated to Java. If memory usage stays well within the allocated maximum memory and performance is still an issue then the performance issues could be elsewhere.

For more information on configuring AppBoard's memory parameters refer to the Memory Configuration page.

CPU

If memory is not an issue but CPU utilization is constantly high then AppBoard isn't able to cope with the load. Java applications automatically make use of multiple CPU cores so allocating additional CPUs, or migrating to a faster/better system will help - or scaling out horizontally (see Scaling AppBoard below).

The CPU load by AppBoard is directly related to demand which depends on how many concurrent users are logged in and what dashboard happen to be active. By using appropriate caching and avoiding processing large data-sets it may be possible to reduce the load on the system.

Also, if AppBoard is sharing the system with other processes then monitoring will be needed to determine what's using the resources.

Responsiveness

Responsiveness can influence the perceived performance of AppBoard by end-users of the system. Clearly the AppBoard server is a factor but in the situation both memory and CPU utilization are not an issue then the problem could be due to other reasons. Consider:

  • slow responses from data sources: such as slow running database queries, high latency links / remote data sources, and poorly performing data sources which is sometimes the case with web APIs.
  • transferring large amounts of data, this applies from the AppBoard server to data sources, and the AppBoard server to clients.
  • Viewer client system doesn't have enough memory
  • Viewer client system is already slow and/or loaded due to other applications
  • Viewer client using large amounts of CPU due to misc reasons such as widgets trying to show too much data, too many widgets, etc...

Some of these can be mitigated to an extent by isolating the problem and trying to reduce or eliminate it. Some examples:

  • slow database query: increase the AppBoard cache timeout if applicable, restructure the query, perhaps indexes may help on the database, enable Server Polling, or handle it external to AppBoard by pre-fetch the data with a batch process and store in a local database which can serve the data much quicker.
  • large data-sets: restructure the query to return the minimum data required. Use server-side-filters to reduce data transferred to the Viewer client.

Refer to the Caching & Polling Overview for more information.

Scaling AppBoard

Scaling AppBoard beyond a single server is supported with the ideal configuration being a load-balanced cluster. As the load on AppBoard is directly related to the demand (concurrent users) then clustering provides a linear way to scale and service additional concurrent users.

For more information please refer to the Clustering & Failover guide.