How Google keeps its database ticking

Google is used to sifting through huge amounts of information to generate its search results, but a 12 gigabyte database proved something more of a challenge for its own financial management and planning systems.

case study Google is used to sifting through huge amounts of information to generate its search results, but a 12 gigabyte database proved something more of a challenge for its own financial management and planning systems.

Since 2001, Google has been using a variety of products from Hyperion to manage its internal financial systems. As the company grew rapidly following its listing in 2004, its approach of using a single database for both forward financial planning and actual financial results become troublesome.

The database grew in size to more than 12 gigabytes, and the period restructuring required to ensure accuracy could see the system, which is now used by more than 150 staff, taken offline for two hours at a stretch.

"The database that we had was already at the limit," Chris Schulze, technical program manager for Google, said during a presentation at Hyperion's Solutions 2007 conference in Orlando.

"There's an inherent conflict having both applications on the same cube," said Chad Coughlin, manager for consultancy Answerthink, which worked with Google to solve the problem.

Rather than ditching the system, Google decided to use partitioning to split the single existing databases into three components: one for the current financial planning projections, one for the actual current data from existing HR and general ledger systems, and one storing historic information.

"We wanted the user experience to be the same, even if it was more complex in the background," Coughlin said. Google also wanted to make it possible to drill down through existing information to more detailed information from its Oracle Financials system.

Using a transparent partition enabled Google to split the three systems while maintaining a single user interface, and built a series of interfaces to enable drill-down. The partitioning process commenced in November 2006, and was completed by February this year.

The forward planning and actual results databases are now each less than one gigabyte in size. The archival database remains relatively large, and can only be fully restructured on weekends as it still takes a number of hours.

Future plans for the system include shifting to Hyperion's System 9 platform and ditching the current Windows server used to run the service.

"Right now, we're on a not very powerful Windows box," Couglin said. "We definitely are wanting to go to Unix when we go to System 9."

Angus Kidman travelled to Orlando as a guest of Hyperion.

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All