Why Is My App SLOw? Defining Reliability in Platform Engineering • Jez Humble • YOW! 2023

Why Your App is Slow: Defining Reliability in Platform Engineering with Jez Humble at YOW! 2023 - Learn how to detect platform latency regressions and improve the customer experience.

Key takeaways

Reliability in platform engineering is about understanding the behavior of workloads and detecting platform latency regressions.
Platform customers care about the customer experience, not the actual behavior of the system.
To measure reliability, you need to understand the normal behavior of your system and detect when it deviates from that norm.
Z-scores can be used to detect anomalies in request delivery latency.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
Probers can be used to detect narrow scenarios, but they are not always effective.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s goal is to make the signal as clean as possible, and to detect when the signal is not clean.
By aggregating Z-scores across workloads, you can get a signal that is combinable across projects and regions.
The platform’s behavior can be complex and unpredictable, and it is difficult to model.
To detect platform latency regressions, you need to understand the normal behavior of your system and detect when it deviates from that norm.
The platform’s

Why Is My App SLOw? Defining Reliability in Platform Engineering • Jez Humble • YOW! 2023

More talks