Explore a novel availability metric for cloud applications in this 24-minute conference talk from NSDI '20. Dive into the concept of "windowed user-uptime" and its significance for both users and system developers at Google's G Suite. Learn how this metric directly models user-perceived availability, avoiding biases found in common availability metrics. Understand the importance of distinguishing between frequent short outages and fewer long-duration outages. Examine the components of user-uptime ratio, including deriving uptime and downtime per user, and visualizing user-uptime across windows. Compare user-uptime to ground truth and success ratio metrics. Investigate the relationship between availability and hyper-active users in cloud applications.
Overview
Syllabus
Intro
Commonly used measure: Success ratio
Commonly used measure: uptime ratio
Why local outages are the norm
User-uptime ratio: uptime from each user's perspective
Deriving uptime and downtime per user
Visualizing user-uptime across windows
User-uptime versus ground truth
Does user-uptime match success ratio?
Availability and hyper-active users
Taught by
USENIX
