Update on the t-digest: Finding Faults in Real Data

06/13/2017 - 11:00 to 11:40
long talk (40 min)

Session abstract: 

Is your system working? Really? Average response times and throughputs don’t tell the whole story. To really understand what is happening, you probably need measurements like the 99.9%-ile response time. A growing number of systems are using the t-digest to do this. I will explain the algorithm with practical examples, talk about how it is much simpler and faster than before, talk about integration in systems like Elastic, Solar and streamlib, tell some real-world deployment stories and show some pretty pictures.