A Few of Our Favorite Insights from Flink Forward 2016

Posted on Oct 11th, 2016 by

The data Artisans team was very much impressed by this year’s Flink Forward speaker sessions, and the speakers delivered tons of detail on Apache Flink® use cases and benchmarks. Here, we’ll share just a small selection of our favorite insights from the presentations.

And remember, all speaker session recordings and slides are available on the Flink Forward website.

  1. Bouygues Telecom, one of the largest telecom networks in France, is running 30 production applications powered by Flink and is processing 10 billion raw events per day. As of Flink Forward 2015, they were live with 5 Flink applications, so we’re looking forward to hearing about their 180 Flink applications in 2017. (All Slides, Talkbouygues-slide-new-image
  2. Netflix has been evaluating Flink to power their Keystone streaming platform and shared some pretty compelling performance numbers from a benchmark of their use case. Deltas in the slide below are measured against Samza, their current production stream processor. (All Slides, Talknetflix-numbers
  3. Uber has developed a solution for on-demand backfilling in order to generate aggregates retroactively, working within Flink’s single watermark-based time-progress model. (All Slides, Talkuber-slide
  4. Alibaba is using Blink, a system based on Flink, to power real-time search results on its e-commerce sites and is running on thousands of nodes in production. (Talk w/ Slides on Screenalibaba-slide-new
  5. New Relic saw Flink outperform both its current, baseline system as well as Spark Streaming in a benchmark; Flink was able to handle 4000% of New Relic’s current production load without issue, at which point the test became bandwidth-constrained. (All Slides, Talknew-relic-numbers
  6. Zalando described how it runs Flink at the core of its next-gen Business Intelligence platform, powering business process monitoring and ETL for the retailer. (All Slides, Talkzalando-architecture
  7. MapR achieved a 250% improvement on the Yahoo! Streaming Benchmark after tuning Flink (running on MapR), clocking in at 10 million transactions per second. (All Slides, Talkmapr-slide
  8. Otto Group has built (and open-sourced) a framework called Flinkspector, enabling users to define unit tests for Flink data flows. As Flink moves to a more prominent position in production environments, tools to ensure reliability and data integrity become critical for enterprises. (All Slides, Talkotto-slide
  9. King, the makers of Candy Crush Saga, shared how streaming analytics is made available to all of its data scientists via an platform called RBea (Rule-Based Event Aggregator), where internal users can deploy simple analytics scripts in a browser UI. (All Slides, Talkking-slide
  10. Bouygues Telecom makes the list twice this year. Presenters Thomas and Amine gave us a peek inside the company’s real-life control center (pictured below), where engineers monitor the health of mobile and fixed-line telephone networks across all of France. And the graphs shown below are powered by–you guessed it–Flink. (Photo from Slide 11)bouygues-telecom-control-center