Now let's go to the Aggregation view by clicking on Aggregation in the navigation bar:


Selecting your data for aggregation works the same way as for the measurement browser, meaning that you'll pick the data using the same selectors in the upper view. However, there are additional parameters which will decide how exactly the aggregation will be performed. Here are the main parameters of the service :


  • intervalSize : It's the range or "bucket size" which RTM will use in order to group up data points. This will indirectly decide how many dots will be included in each series. A very high value will result into a single-dot summary. A very small value will cause for very many dots to be displayed on the chart. By default, the value provided there will be interpreted as milliseconds. Using the string "auto" will cause for the server to compute a value resulting in 30 buckets (configurable).
  • groupby : also called split. RTM expects a key there. By default we'll use the field "name" to split on the measurement names, but you can use any custom attribute 
  • cpu # : this picks the number of tasks and hence maximum CPU cores which will be used to process the next aggregation request
  • partition # : this value represents the number of sub-buckets which will be created to process the data points belonging to each bucket. You'll want to keep this value a multiple of the number of cpus, but you'll need to reduce it in the event that you're computing many dots (high granularity). If you expect that your series will produce hundreds of dots / buckets, make sure to reduce this value to 1
  • timeout (s) : this value will decide how long an aggregation request is allowed to run for. Upon reaching the timeout value, all objects related to that request will be decommitted and garbaged. Increase this value if you're computing a particularly large set, which you expect will take unusually long to be processed

Once you've sent an aggregation request using the Search button, and in the even that the result isn't immediately complete, you'll be able to use the Pause and Resume buttons if you wish for the chart to stop redrawing.


Once the results start streaming back, you can pick the metric you want in the drop down metric menu :



Here's what these metrics means :

  • MIN: minimum value for the interval - exact
  • MAX: minimum value for the interval - exact
  • CNT: number of measurements in the interval - exact
  • SUM: sum of all the values in the interval - exact
  • AVG: mean value in the interval - exact
  • TPS: transactions per second, meaning the number of measurements / interval size x 1000 milliseconds. It provides you with a throughput metric that's independant of the interval size you've picked  - exact
  • TPM: transactions per minute, meaning the number of measurements / interval size x 60000 milliseconds. It provides you with a throughput metric that's independant of the interval size you've picked  - exact
  • PCL50: the median value for the interval (also known as 50th percentile) - approximate (precision can be configured)
  • PCL80: the value of the 80th percentile - approximate (precision can be configured)
  • PCL90: the value of the 90th percentile - approximate (precision can be configured)
  • PCL99: the value of the 99th percentile - approximate (precision can be configured)

You can also decide to turn on the table view (which is switched off by default) by clicking the checkbox in its title:


The table view is great for summaries (i.e single-dot series), whereas a chart doesn't have much value in that scenario.

Created by Jonathan Rubiero on 2018/03/26 12:58
Copyright © exense GmbH