StarZero All shows Try the podcast clipper →

Nov 23, 2015 · 22m · mad

Data: Monitoring to Analytics to Understanding // Alexis Le-Quoc, Datadog

Alexis Le-Quoc · 18m spoken Matt Turck · 25s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

At DataDrivenNYC, Datadog Founder and CTO Alexis Le-Quoc explains how engineering teams can navigate the massive explosion of modern cloud telemetry by pairing automated statistical data science with structured mental models. Through his presentation and Q&A, he details the progression from basic metric collection to advanced analytics and intelligent system understanding.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 2.6% of the talking time here. How this is scored →

Matt as informed peer 0.4 Guest teaching 0.4 Guest disagreement 0.2 Matt pushing back 0.2
05100:0010:0020:000:46–3:20 · Matt as informed peer 0/10 The Evolution of Monitoring and Data Aggregation Alexis opens his presentation introducing Datadog's core functions of data aggregation, visualization, and alerting. The host does not speak during this monologue segment.3:20–7:50 · Matt as informed peer 0/10 Real-Time Alerting and the Quality of Monitoring Alexis outlines the historical progression from bare metal servers to virtual machines and Docker containers, driving exponential metrics growth. The host remains silent throughout the talk.7:50–11:33 · Matt as informed peer 0/10 Information Overload and the Speed of Software Releases Alexis discusses rapid deployment rates creating noise and explains mathematical techniques like outlier and anomaly detection to extract signals. The host does not participate in this presentation block.11:33–15:12 · Matt as informed peer 0/10 Establishing Mental Models: Work Metrics vs. Resource Metrics Alexis explains mental models using an astronomy epicycle analogy and advocates categorizing work metrics versus resource metrics to simplify monitoring. The host remains absent during the monologue.15:12–22:18 · Matt as informed peer 2/10 Summary and Recruitment Pitch Matt enters to host Q&A, asking basic follow-up questions about machine learning models and training requirements, which Alexis clarifies smoothly before taking audience questions.0:46–3:20 · Guest teaching 0/10 The Evolution of Monitoring and Data Aggregation Alexis opens his presentation introducing Datadog's core functions of data aggregation, visualization, and alerting. The host does not speak during this monologue segment.3:20–7:50 · Guest teaching 0/10 Real-Time Alerting and the Quality of Monitoring Alexis outlines the historical progression from bare metal servers to virtual machines and Docker containers, driving exponential metrics growth. The host remains silent throughout the talk.7:50–11:33 · Guest teaching 0/10 Information Overload and the Speed of Software Releases Alexis discusses rapid deployment rates creating noise and explains mathematical techniques like outlier and anomaly detection to extract signals. The host does not participate in this presentation block.11:33–15:12 · Guest teaching 0/10 Establishing Mental Models: Work Metrics vs. Resource Metrics Alexis explains mental models using an astronomy epicycle analogy and advocates categorizing work metrics versus resource metrics to simplify monitoring. The host remains absent during the monologue.15:12–22:18 · Guest teaching 2/10 Summary and Recruitment Pitch Matt enters to host Q&A, asking basic follow-up questions about machine learning models and training requirements, which Alexis clarifies smoothly before taking audience questions.0:46–3:20 · Guest disagreement 0/10 The Evolution of Monitoring and Data Aggregation Alexis opens his presentation introducing Datadog's core functions of data aggregation, visualization, and alerting. The host does not speak during this monologue segment.3:20–7:50 · Guest disagreement 0/10 Real-Time Alerting and the Quality of Monitoring Alexis outlines the historical progression from bare metal servers to virtual machines and Docker containers, driving exponential metrics growth. The host remains silent throughout the talk.7:50–11:33 · Guest disagreement 0/10 Information Overload and the Speed of Software Releases Alexis discusses rapid deployment rates creating noise and explains mathematical techniques like outlier and anomaly detection to extract signals. The host does not participate in this presentation block.11:33–15:12 · Guest disagreement 0/10 Establishing Mental Models: Work Metrics vs. Resource Metrics Alexis explains mental models using an astronomy epicycle analogy and advocates categorizing work metrics versus resource metrics to simplify monitoring. The host remains absent during the monologue.15:12–22:18 · Guest disagreement 1/10 Summary and Recruitment Pitch Matt enters to host Q&A, asking basic follow-up questions about machine learning models and training requirements, which Alexis clarifies smoothly before taking audience questions.0:46–3:20 · Matt pushing back 0/10 The Evolution of Monitoring and Data Aggregation Alexis opens his presentation introducing Datadog's core functions of data aggregation, visualization, and alerting. The host does not speak during this monologue segment.3:20–7:50 · Matt pushing back 0/10 Real-Time Alerting and the Quality of Monitoring Alexis outlines the historical progression from bare metal servers to virtual machines and Docker containers, driving exponential metrics growth. The host remains silent throughout the talk.7:50–11:33 · Matt pushing back 0/10 Information Overload and the Speed of Software Releases Alexis discusses rapid deployment rates creating noise and explains mathematical techniques like outlier and anomaly detection to extract signals. The host does not participate in this presentation block.11:33–15:12 · Matt pushing back 0/10 Establishing Mental Models: Work Metrics vs. Resource Metrics Alexis explains mental models using an astronomy epicycle analogy and advocates categorizing work metrics versus resource metrics to simplify monitoring. The host remains absent during the monologue.15:12–22:18 · Matt pushing back 1/10 Summary and Recruitment Pitch Matt enters to host Q&A, asking basic follow-up questions about machine learning models and training requirements, which Alexis clarifies smoothly before taking audience questions.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 7.7% · guest 92.3%15:00 · Matt 7.7% · guest 92.3%18:00 · Matt 11.4% · guest 88.6%18:00 · Matt 11.4% · guest 88.6%21:00 · Matt 2.6% · guest 97.4%21:00 · Matt 2.6% · guest 97.4%
Sharpest disagreement ▶ 20:35 Playful joke about intimidating the audience

In a completely collaborative talk, Alexis makes a lighthearted joke that he scared the audience when no hands go up immediately for questions.

 Hardest push from Matt ▶ 16:37 Matt presses on training time requirements

Matt asks a direct operational follow-up inquiring how quickly anomaly detection works and how much training data is required.

 Biggest teaching moment ▶ 15:44 Alexis clarifies machine learning vs signal processing

When Matt asks if anomaly detection is simply running machine learning, Alexis reframes the concept by explaining the difference between classic statistical signal processing and ML clustering.

 Matt holds his own ▶ 15:44 Matt frames anomaly detection around machine learning algorithms

Matt demonstrates technical understanding of the presentation by correctly linking statistical data science concepts to underlying machine learning processes.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
The Evolution of Monitoring and Data Aggregation 0000 Alexis opens his presentation introducing Datadog's core functions of data aggregation, visualization, and alerting. The host does not speak during this monologue segment.
Real-Time Alerting and the Quality of Monitoring 0000 Alexis outlines the historical progression from bare metal servers to virtual machines and Docker containers, driving exponential metrics growth. The host remains silent throughout the talk.
Information Overload and the Speed of Software Releases 0000 Alexis discusses rapid deployment rates creating noise and explains mathematical techniques like outlier and anomaly detection to extract signals. The host does not participate in this presentation block.
Establishing Mental Models: Work Metrics vs. Resource Metrics 0000 Alexis explains mental models using an astronomy epicycle analogy and advocates categorizing work metrics versus resource metrics to simplify monitoring. The host remains absent during the monologue.
Summary and Recruitment Pitch 2211 Matt enters to host Q&A, asking basic follow-up questions about machine learning models and training requirements, which Alexis clarifies smoothly before taking audience questions.

Statements from this episode (8)

Assertion Not checkable as stated
Le-Quoc: Vast majority of industry still struggles with data visualization
“Not the sort of advanced customer of ours, you know, say, you know, Airbnb or Twitter or, Spotify, but the vast majority of the industry is, is still kind of struggling with that.”
Alexis Le-Quoc Nov 23, 2015 ▶ 2:48
Insight
Le-Quoc: Continuous dashboard monitoring fails due to human visual fatigue
“Once you can put it on a screen, it's great, but you're not gonna watch watch this for an entire day, or even a few hours, because A, you'll get tired of it, B, your brain will sort of things will become blurry, and you'll, you'll use your, you'll lose your ef…”
Alexis Le-Quoc Nov 23, 2015 ▶ 3:04
Insight
Le-Quoc: A team's software understanding is only as good as its monitoring
“For all the customers the understanding that they have of the application is only as good as their monitoring.”
Alexis Le-Quoc Nov 23, 2015 ▶ 4:32
Insight
Datadog CTO: Software uniquely rewards maximizing the rate of deployment change
“In the industry of software, we pay a premium for rate of change. By that I mean, and it's very different from a physical production line, is the faster the production line changes, the better off we are. That means the faster you can go to market the more fea…”
Alexis Le-Quoc Nov 23, 2015 ▶ 8:20
Insight
Le-Quoc: Monitor statistical outliers rather than individual metric time series
“I don't want to monitor any particular of the time series, because there are too many of them, just the volume of data is too large. Rather, I want to find the ones that, that sort of deviates.”
Alexis Le-Quoc Nov 23, 2015 ▶ 10:46
Insight
Alexis Le-Quoc: Prioritize work metrics over resource metrics in system monitoring
“Not all metrics are created equal, and the ones you want to really care about it are the work metrics. So in that large volume of metrics, if you already classify in, in two classes of metrics, work and resource, you can reduce your search space by a lot.”
Alexis Le-Quoc Nov 23, 2015 ▶ 14:27
Insight
Datadog CTO: False positives, not missed alerts, are the primary enemy in monitoring
“So that, that's why I would say for us to, it's really false positives. That's the enemy. Somehow you'll find out when something goes wrong.”
Alexis Le-Quoc Nov 23, 2015 ▶ 19:44
Assertion Not checkable as stated
Datadog data showed an average of four Docker containers per host
“We actually looked at our data set for adoption, and so things we extracted are number of containers on average, or distribution of containers per machine, which I think the average was about four”
Alexis Le-Quoc Nov 23, 2015 ▶ 21:19
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more