King.com has over 40 million registered players and offers games online and via mobile devices, where it gets over half of its traffic through Facebook.
The company's director of data warehousing, Mats-Mats Eirksson, told ComputerworldUK that analytics is vital to its success online.
"Analytics is one of the things that made king.com the thing that it is today. In the universe that we operate in, gaming online, it is absolutely essential to know as much as possible about the players and optimise everything," said Eriksson.
"You can get so much information just by looking at the user behaviour, which you can then use to create better games and better monetisation features. At the end of the day we want to make money out of this."
King.com is replaced its Infobright platform with Hadoop at the end of last year. Hadoop uses a process called parallel programming, which allows analytics to be run on hundreds of servers, with lots of disk drives, all at the same time. It stores this data in a file system called HDFS (Hadoop distributed file system), in effect a flat file system that can spread data across multiple disk drives and servers.
It is generally agreed that Hadoop is an extremely complex system to master and requires intensive developer skills. There is also a lack of an effective ecosystem and standards around the open-source offering.
King.com uses the technology to analyse 'events', which is any action taken within a game by a player. This could be the starting of a game, any transaction made, attempts tried at each level, sharing habits, all of which are written into a line in the text file and stored on the web server log file. The log files from every server are then moved into HDFS and the Hadoop environment for analysis.
"We needed something that could scale because we saw an almost exponential increase in the number of events we recorded from our games, which was driven by us increasing the number of games we offer, but also by increasing the variety of events we recorded," said Eriksson.
"Everybody wants a business case for Hadoop, but for me it is simply about difference between knowing what happens in a game and not knowing For example, one thing we can analyse that requires a substantial amount of data is game optimisation," he added.
"Let's say that we have 150 levels that the user will complete one by one. If for whatever reason level 50 is too hard, then people will be stuck at that point and get tired of the game. So, we need to make sure that the difficulty increases very smoothly across the user progression. It's about fine tuning the game."
King.com has nine computers running its Hadoop cluster and nine analysts looking the data. However, Eirksson said that this is a 'moving target' and plans to increase capacity often.
He also said that finding the skills to run Hadoop, which can be tricky, wasn't a problem for the gaming giant.
He said: "We were lucky because the founders of King have a lot of connections in the IT world and got a contractor from a start-up called Big Data AB.
"He has been working with us for a year now and without his knowledge and great engineering skills it would have been quite hard to do because nobody internally knew the best way to set this up."
He added: "It can be hard to find people to work with Hadoop because you have to be a cross-breed between data warehousing, analytics and programming."