In June 2016, (part of) the growing Rigor team attended speaker sessions at the O’Reilly Velocity conference in Santa Clara. We attended sessions lead by performance and DevOps experts like Ilya Grigorik, engineering researcher and physician Richard Cook, and even our own Web Perfectionist, Billy Hoffman. The three day conference was full of education, thought provoking discussions, and networking. In case you weren’t able to attend, here are takeaways from some of our favorite speaker sessions.


1. “Building a user-centric ops, support, and engineering team” – Peter van Hardenberg, Heroku

Summary:
How to build a highly scaled organization with world-class operations and support and a deep appreciation for the challenges users face.

Takeaways:
User-centric teams should:

  • Routinely analyze customer support issues to ensure that customer pain radiates through the organization.
  • Review on-call issues each week to diagnose recurring issues, identify alert fatigue, and reduce pager burden.
  • Celebrate retiring old code and services (via “burn parties”) to encourage cleanup and reward traditionally thankless work.
  • Ensure all teams have an end-user, and that all roles needed to deliver value to the end-user are a part of each team.

Putting the talk into action…
After hearing this talk, Rigor Engineering wants to schedule our first “burn party”. We’ve upgraded or retired a fair amount of legacy components in the past year, improving our velocity and boosting developer happiness. It’s important to celebrate this work as it’s often invisible to the casual observer.

Summary | Slides

 

2. “The wild west of media performance: A Vox story” – Ian Carrico and Jason Ormand, Vox Media

Summary:
The story of Vox Media’s creation of a dedicated performance team and initiative to completely revamp their product with the goal of having the fastest sites possible.

Takeaways:

  • Testing and tracking performance can require multiple complimentary software solutions. Use both RUM and Synthetic Monitoring.
  • Defining a methodology for testing makes performance analysis repeatable and scalable. (Vox’s methodology included in slide deck)
  • Implement page asset budgets to track performance regression. Check out Vox’s open source tool or use Rigor to establish performance budgets.
  • Implement image quality control process across the organization. Investigate image formats like WebP and leverage preload for critical images.


If we could do one thing…
Take the methodology and process that Vox’s Product team used in this example and implement it across other online media organizations who want to tackle performance on legacy systems.

Summary | Slides

 

3. “Using machine learning to determine drivers of bounce rate and conversion” – Patrick Meenan, Google and Tammy Everts, SOASTA

Summary:
SOASTA and Google partner to run a machine learning model on anonymous RUM data and determine the strongest predictors of bounce and conversion rate for e-Commerce sites.

Takeaways:

  • Sessions that converted had 38% fewer images than sessions that didn’t.
  • DOM ready was greatest indicator of bounce rate.
  • Full load time was second greatest indicator of bounce rate.
  • Out of 93 conventional metrics “Start render” and “DNS lookup” rank 69 and 79 respectively.
Bounce rate performance predictors for SOASTA/Google E-Commerce machine learning study.
Bounce rate performance predictors for SOASTA/Google E-Commerce machine learning study.

If we could do one thing…
Gather Rigor data from our users and run a similar analysis to determine if our findings are consistent. This would help us give our customers more insight as to what performance metrics to zone in on. We’d be interested to run the same machine learning from RUM and Synthetic data gathered from online media companies.

Summary | Slides

What were your favorite talks from Velocity? Are you attending Velocity in New York or Amsterdam? Let us know in the comments below or on Twitter @TeamRigor, we’d love to talk!