How much money would you save if AI could detect and fix your outages as soon as they happen? In a multi-billion dollar business, outages are very expensive. MTTR has a direct effect on the bottom-line, so every second count in resolving issues. But with millions of metrics being generated by thousands of microservices, how do you choose which metrics to pay attention to? How do you make your alerts meaningful to avoid alert fatigue and desensitisation? How do you respond to those alerts in a timely manner?

In this talk, Matt covers how Expedia is using Machine Learning to "close the loop" involved in detecting, diagnosing and remediating outages post-release. You will learn about how to use ML to build models for anomaly detection in metrics. You will also learn about "ML-Ops" and how to build a platform for training and deploying ML models.

How much money would you save if AI could detect and fix your outages as soon as they happen? In a multi-billion dollar business, outages are very expensive. MTTR has a direct effect on the bottom-line, so every second count in resolving issues. But with millions of metrics being generated by thousands of microservices, how do you choose which metrics to pay attention to? How do you make your alerts meaningful to avoid alert fatigue and desensitisation? How do you respond to those alerts in a timely manner?

In this talk, Matt covers how Expedia is using Machine Learning to "close the loop" involved in detecting, diagnosing and remediating outages post-release. You will learn about how to use ML to build models for anomaly detection in metrics. You will also learn about "ML-Ops" and how to build a platform for training and deploying ML models.

How much money would you save if AI could detect and fix your outages as soon as they happen? In a multi-billion dollar business, outages are very expensive. MTTR has a direct effect on the bottom-line, so every second counts in resolving issues. But with millions of metrics being generated by thousands of microservices, how do you choose which metrics to pay attention to? How do you make your alerts meaningful to avoid alert fatigue and desensitisation? How do you respond to those alerts in a timely manner?

In this talk, Matt covers how Expedia is using Machine Learning to "close the loop" involved in detecting, diagnosing and remediating outages post-release. You will learn about how to use ML to build models for anomaly detection in metrics. You will also learn about "ML-Ops" and how to build a platform for training and deploying ML models.

We know the “right thing” to do is to work together, to remove bottlenecks, to automate, automate, automate. But when the status quo is finger-pointing, mistrust, blame, and over-exhaustion from increasingly complex manual releases, it seems impossible to find a way to climb out of the downward spiral. How did an IT department turn this around to become a harmonious department with common goals and over 10x faster cycle times?

Winning the hearts and minds needed for lasting DevOps change requires something more than just great automation. This is the story of how Wotif found a way to incentivise a DevOps transformation across an entire IT department, resulting in cycle times measured in hours instead of weeks/months. While this journey involved technology such as DropWizard, Puppet & ZooKeeper, Matt (dev) and Alexandra (ops) extract principles that they hope can be applied to any organisation at grassroots and leadership levels using existing toolchains to support not only the best ideas of the present but also the legacy of the past and the unknown innovation of the future.

How much money would you save if AI could detect and fix your outages as soon as they happen? In a multi-billion dollar business, outages are very expensive. MTTR has a direct effect on the bottom-line, so every second counts in resolving issues. But with millions of metrics being generated by thousands of microservices, how do you choose which metrics to pay attention to? How do you make your alerts meaningful to avoid alert fatigue and desensitisation? How do you respond to those alerts in a timely manner?

In this talk, Matt covers how Expedia is using Machine Learning to "close the loop" involved in detecting, diagnosing and remediating outages post-release. You will learn about how to use ML to build models for anomaly detection in metrics. You will also learn about "ML-Ops" and how to build a platform for training and deploying ML models.

We know the “right thing” to do is to work together, to remove bottlenecks, to automate, automate, automate. But when the status quo is finger-pointing, mistrust, blame, and over-exhaustion from increasingly complex manual releases, it seems impossible to find a way to climb out of the downward spiral. How did an IT department turn this around to become a harmonious department with common goals and over 10x faster cycle times?

Winning the hearts and minds needed for lasting DevOps change requires something more than just great automation. This is the story of how Wotif found a way to incentivise a DevOps transformation across an entire IT department, resulting in cycle times measured in hours instead of weeks/months. While this journey involved technology such as DropWizard, Puppet & ZooKeeper, Matt (dev) and Alexandra (ops) extract principles that they hope can be applied to any organisation at grassroots and leadership levels using existing toolchains to support not only the best ideas of the present but also the legacy of the past and the unknown innovation of the future.

We know the “right thing” to do is to work together, to remove bottlenecks, to automate, automate, automate. But when the status quo is finger-pointing, mistrust, blame, and over-exhaustion from increasingly complex manual releases, it seems impossible to find a way to climb out of the downward spiral. How did an IT department turn this around to become a harmonious department with common goals and over 10x faster cycle times?

Winning the hearts and minds needed for lasting DevOps change requires something more than just great automation. This is the story of how Wotif found a way to incentivise a DevOps transformation across an entire IT department, resulting in cycle times measured in hours instead of weeks/months. While this journey involved technology such as DropWizard, Puppet & ZooKeeper, Matt (dev) and Alexandra (ops) extract principles that they hope can be applied to any organisation at grassroots and leadership levels using existing toolchains to support not only the best ideas of the present but also the legacy of the past and the unknown innovation of the future.

We know the “right thing” to do is to work together, to remove bottlenecks, to automate, automate, automate. But when the status quo is finger-pointing, mistrust, blame, and over-exhaustion from increasingly complex manual releases, it seems impossible to find a way to climb out of the downward spiral. How did an IT department turn this around to become a harmonious department with common goals and over 10x faster cycle times?

Winning the hearts and minds needed for lasting DevOps change requires something more than just great automation. This is the story of how Wotif found a way to incentivise a DevOps transformation across an entire IT department, resulting in cycle times measured in hours instead of weeks/months. While this journey involved technology such as DropWizard, Puppet & ZooKeeper, Matt (dev) and Alexandra (ops) extract principles that they hope can be applied to any organisation at grassroots and leadership levels using existing toolchains to support not only the best ideas of the present but also the legacy of the past and the unknown innovation of the future.