JMertin

TIM performance juxtaposed

Blog Post created by JMertin Employee on Sep 19, 2017

having had the cemperfdb.ca.com site available for some time now, it took me a while to find the time to actually use the data and put it in comparison to other running systems.

For those not knowing the CEM Performance database (this is essentially meant for people outside of the CA Lan), this is a UI displaying performance data and visualizing data as collected by the TIM in form of protocolstats. All that is required is to collect the protocolstats of the TIM system using the https://github.com/CA-APM/cem-healthcheck-scripts  / TIMPERF data collection run, send it to one CA Support/SWAT member for upload, and voila - a full performance report of that system + visualization over time of the collected data is available.

 

Up to now however, all the comparison between different systems had to be done manually bringing our brains to run hot. Recently, a new functionality has been added enabling us to see the selected system placed in a percentile based table, while the data from other systems of the same type are used as "reference" data.

In the end, this is how it looks like:

 

Please take the following points into consideration for reading the percentile distribution table:

  • By default, if a host for comparison exists, the system will automatically compare to the same hardware only
  • The highest concentration of hosts will be highlighted using a Red Color frame.
  • The selected host for comparison will have its location in the table highlighted in Orange Color.
  • The median values are computed using the average values used for the percentile calculations.
  • 0 values are ignored for Average and Median calculations
  • 0 values are not taken into account. This can lower the count of systems per line as per valid data sets.
  • In case network throughput remains 0 or the sum of the calculated stats is 0, the entire system is ignored.
  • All average data computations are limited to working hours only (08h00 -> 17h00).
  • 0.001% of the highest values have been cut out to remove the existing peaks which cause a false percentile perception
  • In case no Orange Color value field for the host to compare in a lign exists, it just means that this host has values out of the range of the percentile table. This only happens if the host to be compared is compared to lower performing hardware class.

 

Note - if anything not clear, let me know what needs to be changed. It took quite some time to find a balance between real data, and the peaks due to overloaded systems, too low average as night data went down to 0 etc.

 

How do I access this view on the cemperfdb.ca.com UI?

Once the host-data is uploaded and analyzed, use the highlighted link to go to the percentile situation of the current host.

Outcomes