GoDaddy at Velocity 2016 New York

Copyright© 2016 GoDaddy Inc. · 14455 N. Hayden Road Scottsdale, Arizona 85260 (480) 505-8800 · All Rights Reserved.

How GoDaddy radically improved web performance and increased customer engagementJim Pierson

Velocity NY - Sept 22, 2016

CONFIDENTIAL. COPYRIGHT © 2016 GODADDY INC. ALL RIGHTS RESERVED.2Copyright© 2016 GoDaddy Inc. · 14455 N. Hayden Road Scottsdale, Arizona 85260 (480) 505-8800 · All Rights Reserved.

The Money slide• On June 4th, 2015, GoDaddy deployed its

fastest India home page.

• The next day, user purchases shot up several percentage points. Soon after, we rolled out the solution to all of our sales sites.


Agenda• How GoDaddy matured as a Performance organization

• The Tools and Best Practices that worked

• How Big Data collection and Dashboards changed the game

• Anomaly detection in Production and Testing, it’s a math problem.

• The need to Join forces

Performance Engineering Maturity Level 1 - Novice

Novice

Proactive

Reactive

• Initial :• Firefighting, chasing• Blind, apathy, lacking TRUST• Hands-On learning of the tools

• Eventual:• Investigative approach

• Trench coat vs Lab coat• Look at every angle

• Tool Proficiency• Patterns evolve into Best Practices

• Exit:• Measure Everything -> RUM, APM

• Relapse:- too many new toys


Top Fixes – Deferring JS• Defer JavaScript:

• UXCORE

• Tag Mangers

• LIVE CHAT


Top Fixes – Deferring JS


Top Fixes – File size optimization• File size reduction:

• Combine JS/CSS

• Minify JS/CSS

• Compress everything!

• Review/optimize:• Images

• Fonts

• CSS

• JS


Top Fixes – Head Flushing• HEAD flushing

(chunking)• Fetch CSS

immediately!


UglyNot so good PLT• Onload: 5s in Phx, 13s in HK

• VisComplete: 4.8s in Phx, 6.3s in HK

• 13s in Hong Kong

• 133 items

• Nothing deferred

• No CDN Domain

http://wpt.intranet.gdg/result/160712_W4_256/1/details/

....


Nice!A good site PLT• Onload: 1.6 seconds in Phx, 1.7 in HK

• VisComplete: 2.5s in Phx, 3.8s in HK

• 14 items b4 onload

• Faster Server Response Time – Head Flushing

• JS Deferred

• Tags Deferred

• Initial images limited

• Minimized css, Fonts

• Oops, one css after Onload, reflow

http://wpt.intranet.gdg/result/160712_4Q_255/1/details/ (Phx)http://www.webpagetest.org/result/160712_AT_18DC/1/details/ (HK)

http://wpt.intranet.gdg/result/160712_4Q_255/1/details/

http://www.webpagetest.org/result/160712_AT_18DC/1/details/

Goal oriented

Novice

Performance Engineering Maturity Level 2 – Goal oriented

Proactive

Reactive

• Initial :• Aspirational goals• Limited storage, visualization• Data not JOIN’d and not defined

• Eventual:• Storing everything• Everyone is using the dashboard

• Exit:• Regular Reviews and faster troubleshooting

• Relapse:• Too lenient on ship goals• Not watching the changes

Measure Everything


Real User Monitoring• Browser based

• Hadoop, Elastic

Search, Kibana (ELK)

• Test in production


RUM Scorecard for Execs


CDN Measurement with RUM – Resource TimingCedexis uses a RUM tag added to our page to measure CDN perf


Application Performance MonitoringNew Relic APM agentFaces 2 directions: (transactions, and external transactions)


Using APM to make comparisons


Keep-Alives We are reopening TCP on every transaction

8 RTs x 60ms = 480ms

Goal oriented

Novice

Performance Engineering Maturity Level 3 – Under Control

Proactive

Reactive

• Initial :• 3 steps forward, 1 step back• Cant get at the data fast enough (hours)

• Eventual:• Dynamic threshold math• Inline anomaly detection for production• Catching bad perf before production

• Exit:• Cross-Org communications

• Relapse:- Alerting complexities- Improper hand-off of Alerts

Measure Everything

AutomateEverything


Anomaly detection - Change Index


Change Index - Rank


Real User Monitoring - Alerting


Perf Tests that can stop the Build• Assert on perf

• Using RUM baseline

• View results in ELK

dashboard

Goal oriented

Novice

Performance Engineering Maturity Level 4 – Best of Breed

Proactive

Reactive

• Initial :• Everyone is involved, no one owns it• Super complex• Putting off for years major architectural issues

• Eventual:• Everyone bought into the big event • and in lock-step

• Exit:• 3-5 year arc per product group

• Relapse:- New products

Measure Everything

AutomateEverything

Communicate With Everyone


Team work and influences

platform

Design/PO

editing

Mid tier

Backend

Legacy

CDN

• Legacy owner can make simple but quick changes

• Platform needs lots of coordination


Pealing back the Onion• Many teams• Many priorities• No clear owner

Cart

Front End

Ads

Auth

Backend

API

CDN

Performance Engineering Maturity Levels - Recap

L4) Joining Forces, Driving Architecture refactoring

Goal oriented

Novice

Measure Everything

CommunicateWith Everyone

Proactive

Reactive

Automate Everything

L2) Low hanging fruit, Best Practices,setting goals, basic monitoring

L1) Firefighting, following the trail, discovery

L3) Anomaly detection - Operational AlarmsControlling regression with CICD


Q & A

[email protected]

GoDaddy at Velocity 2016 New York

Data & Analytics