Webdev:Velocity 2009 Notes
From MozillaWiki
- Metrics that Matter
- Measure end user response time, not network or server time
- Measure flows (registration, checkout, etc)
- Fixing Twitter
- Make everything a queue
- Use google analytics for outage page to track problems
- Shopzilla
- Increased performance has a direct correlation with higher revenue and pageviews.
- Averages don't mean anything, get the median and percentiles.
- Optimization (js)
- Running code is slower than not doing anything
- be aware of lookup chains, use locals
- beware of live collections
- avoid the reflow
- writing javascript like it's C makes it faster
- Facebook
- "Test with real users, don't test in the dark." -- WTF (talking about unit testing, not scaling)
- Flickr
- http://www.kitchensoap.com/2009/06/23/slides-for-velocity-talk-2009/
- automated infrastructure
- shared version control between webdev and IT
- one-step build *and* deploy
- continuous integration server (Hudson)
- dark launches
- ganglia monitoring
- New code (dev)
- what metrics will changes, and how!
- what are the risks, signs something is going wrong
- Ops
- be transparent, give devs access to systems
- Overall
- Graphs or it didn't happen
- Measure everything you can
- IT & Webdev/engineering should overlap and have same goals