Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 387 - Software
Type: Contributed
Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
Sponsor: International Chinese Statistical Association
Abstract #318911
Title: Statistical Methods of Percentile Estimation and Inference for Large-Scale Data in Real Time
Author(s): Tianhong He* and Siteng Hao* and Meeyoung Park
Companies: Google and Google and Google
Keywords: non-parametric; large-scale data; real-time; performance; storage

Quantiles for latency data are well-established and broadly used metrics for monitoring website performance. As the Google Cloud Platform (GCP) is the gateway for all Cloud users to access Google Cloud service, the page performance is critical for successful user journeys. However, the extremely large traffic volume poses a couple unique challenges when we try to conduct real-time latency quantile comparison in online A/B experiments, such as storage limitation and computation time. In this talk, we will discuss how we have developed and implemented statistical methods which significantly reduce the computational cost and accommodate different storage strategies into the GCP performance monitoring system.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program