Our business in Europe continues to expand and I’m excited to share this guest blog post from Geoff Cleaves, Business Intelligence Manager at Billy Mobile a new Hortonworks customer based in Barcelona, Spain.
This week at Billy Mobile we are migrating our core technology stack onto HDP 2.3 and boy are we looking forward to it. Billy Mobile is a new mobile ad exchange based in Barcelona and we’ve been struggling to handle our growing volumes of data. Displaying the right ad to millions of visitors a day in a few milliseconds has been a challenge using PHP, MySQL and Elasticsearch. We’ve chosen Hortonworks as our partner on the road to a billion visits a day.
A high performance platform for a high demand business
Billy Mobile has to deliver excellent eCPMs (effective cost per thousand advertising impressions) to our publishers. Our advertisers demand high-quality users. In order to achieve these requirements, we have to churn through tens of millions of records a day to determine the right offer for the right user using semantic analysis and machine learning. We’re placing our bets on a combination of Kafka, Storm, Hive, Hbase, Spark, and Hortonworks support to get the job done. The visitor’s profile and the displayed ad data is stored in Kafka while Storm streams the information into Hive and Hbase. A batch Spark process then analyzes the Hive data and places a new decision making algorithm on our web front-ends. We use Hbase to achieve low latency reads and writes to quickly match visit information with the events that our advertisers trickle to us. In future we will expand the use of Hbase, in combination with Hive, as the source of webmaster and advertiser information panels as well as the source of an OLAP cube data structure.
In the Hadoop sweet spot
Apache Hadoop’s raison d’être is to handle data velocity in a scalable way and that is what we needed at Billy Mobile. Hadoop has established itself in this realm. We know that when we make the jump to billions of daily impressions Hadoop can handle the load by expanding the cluster. We are further comforted with our decision because Hadoop runs (mostly) on Java and Java developers are in abundance. HiveQL is similar enough to SQL that our business analysts can easily handle the syntax. We can connect Tableau straight to Hive.
Partnering with Hortonworks
It was quite easy for us to decide on Hortonworks as a platform and for support. The 100% open source nature of the platform means we can rely on all the pieces being there, always. Hortonworks has convinced us that we can rely on their support not only for situations where some component isn’t working as expected, but also on a consultative basis when planning our cluster’s growth and integrations.
We expect a bright future for Billy Mobile once we migrate fully onto Hortonworks.
You can read the press release announcing Billy Mobile as a customer here.