Overview
Explore advanced techniques for accurately calculating latency Service Level Objectives (SLOs) in this 50-minute conference talk from SREcon19 Asia/Pacific. Delve into three distinct approaches for precise latency SLO calculations, learn how to leverage histograms for mathematically correct quantile computations, and discover the importance of proper SLO framing. Gain insights on using tools like libcircllhist in various programming languages, understand why traditional metrics like median and percentiles can be misleading, and explore the iterative nature of SLO development. Enhance your understanding of SLO best practices and their critical role in maintaining service reliability.
Syllabus
Intro
Service Level Objectives
A more dramatic example
Rethinking computing SLO latency
Compute the SLO from stored data
Count the number of bad
Using histogram latency data
libcircllhist: C, Java, Go, Javascript
Framing SLOs as quantiles is backwards
Framing SLOs correctly is important
Framing SLOs is iterative
SLOs the undiscussed problem
Taught by
USENIX