In early 2016 we embarked on a journey to help measure and improve the performance of ePledge through the use of industry standard tools. Our primary tools were CA Blazemeter, Apache JMeter, New Relic and Google Analytics which helped to simulate and record ePledge performance characteristics during our stress tests. We shared these results at the Andar Users Conference in Tampa. Google Analytics has also been running against many of our production environments for a few years giving us historical data to compare.
While stress testing our systems we tuned the configuration of Andar ePledge as well as our own infrastructure, which resulted in a dramatic boost in performance. Changes made were a combination of enabling caching of web notes/options, utilizing IIS caching, and increasing the number of ePledge instances. We also identified areas of improvement for Helix and escalated those as projects to them, some of which were included in the last Andar release. Having fast page load times directly affects the overall donor experience and with our changes in place we saw a dramatic reduction in page load times. As this was synthetic testing, we looked forward to seeing what impact this would have on real campaigns this season.
Comparing August 21st through October 31st 2015 with the same 3-month period in 2016 for one United Way, we found mean page load times decreased by 46.74% year over year (5.33 seconds vs 2.84 seconds) which helped improve conversion rates by 9.17%. As these changes were applied for all United Ways hosting ePledge with Upic, we would expect similar improvements for all. Feedback from a company who ran a campaign with significant performance issues last season was very positive this year, further validating the changes.
Heading into this new year we will see Helix requiring a newer version of Tomcat, along with further performance enhancements we have requested of Helix; how will this effect performance? We can’t wait to find out! If you need help improving the performance of Andar ePledge, contact us today.
Acknowledgements
We would like to thank CA Blazemeter (related case study) and New Relic for donating the use of their tools for this project, and all the volunteers who work on Apache JMeter.