Real-Time Ecommerce Analytics at Gilt Group

Notes from "Real-Time Ecommerce Analytics at Gilt Group" from MongoSF
  • Real-time analytics is a sweet spot for MongoDB
  • Gilt gets a ton of traffic at noon, that's when their sales start.
  • "None of our items page, because our users like to scan through"
  • "Things sell out, we have limited depth."
  • How do we improve the conversion of our gifts section?
  • Capture data in mongo
  • Analyze w/ Map Reduce
  • Update TXN systems
  • Repeat
  • Takes 30-35 minutes, mostly running map-reduce... we didn't want every visit to the page to be different... so we ran every 30 minutes
  • "We put Mongo on the server in december, haven't logged into the server since"
  • Pageview sends analytics info via AJAX
  • "Good failure mode" everything works except we don't get analytics
  • We just put into Mongo, if it's too slow, we'll add more nodes and run map reduce over the nodes
  • "Nothing production should hit a relational database at Gilt"
    • Key transaction systems run on Voldemort
  • Map-Reduce is waaay easier in Mongo than it is in Hadoop.  Running in Map-Reduce in mongo is 1-2 seconds.
  • "Relational databases for transactional systems don't work."
  • Node.js mongo
    • var mongo = require("lib/mongodb")
  • All node and mongo is run on a single EC2 server and runs about 2% of CPU