The company's engineers have also updated a Hadoop workflow scheduler called Oozie and Pig, a high-level programming environment for running MapReduce jobs.
Yahoo started dedicating large amounts of engineering work to refining Hadoop around 2006, said Doug Cutting, a search specialist who created Hadoop. "Yahoo had lots of interesting data across the company that be correlated in various way, but it existed in separated systems," said Cutting, who now works for Hadoop distribution provider Cloudera. Hadoop promised an easy way for Yahoo to do cross-system analysis of data.
Cutting built Hadoop as part of a part-time project building open source search software (he named the technology after his son's stuffed elephant). He knew he needed distributed servers given the enormity of the project.
"We realized that doing reliable distributed software was really hard," Cutting said. "Every step in which anything is distributed, there are myriad ways in which it could fail. You have to think about how to handle each failure successfully."
Hadoop offers a unique tool in some circumstances, said Curt Monash of Monash Research. "Hadoop is a great tool for organizing and condensing large amounts of data before it is put into a relational database," he said.
It is also a good tool for companies to analyze relationships between people or things, a practice often known as "social graph analysis," Monash said. "Traditional relational databases have a difficult time with this, because each hop along the graph exponentially increases the amount of work that needs to be done," he said.
But there are tradeoffs with the technology. For one, you may not want to use it for real time data analysis.
Cornelius admits Hadoop has latency issues. Because of its distributed nature, Hadoop is not as fast as other BI systems. But, Cornelius and others argue that Hadoop should not be considered as an alternative to a transactional database system or a data warehouse, but rather something that can do tasks that these technologies would struggle to execute.
"It's not a database. It's a different kind of data storage and analytics platform. If you have a relational database problem, you should go buy Oracle or DB2," agreed Mike Olson, Cloudera CEO. To better pursue the BI market, Cloudera has forged partnerships with Pentaho and data warehouse provider Teradata.
"If you want to combine complex unstructured data from multiple sources, if you want to do sophisticated pattern detection, then Hadoop is your only choice," Olson said.
IDG News Service U.S. correspondent Chris Kanaracus contributed to this report.
