Leave it to Google to make business data processing—among the stodgiest topics in the rather buttoned-up world of database systems—seem cool. The application here involves producing reports over Google's ads infrastructure: Google executives want to see how many ads each Google property is serving, and how profitable they are, and Google's customers want to see how many users are clicking on their ads, how much they are paying, and so on.
At a small scale, solving this problem is straightforward—new ad click and sales data are appended to a database file as they are sent from the processing system, and computing the answer to a particular query over the data involves reading the contents of ("scanning") the data file to compute a running total of the records in the groups the user is interested in. Making this perform at the scale of Google Ads, where billions of clicks happen per day, is the challenge addressed by the Mesa system described in this following paper.
No entries found