Microsoft completes Hadoop U-Turn, trumpets HDInsight on Azure

Microsoft has spun up a Hadoop service on its Azure cloud, capping off a tumultuous few years in which Redmond first competed against the open-source technology, then came round to it.

The arrival of HDInsight as a general availability product on Azure was announced by Microsoft today, coming after the data muncher'n'cruncher spent a year in preview on the service. It gives admins access to a capable data analytics platform that can be bought on a per-hour basis.

"Our customers are demanding more from the data they have – not just higher availability, global scale and longer histories of their business data, but that their data works with business in real time and can be leveraged in a flexible way to help them innovate," the Windows giant wrote.

Microsoft had previously developed its own rival to Hadoop named Dryad, though killed off development of this in 2011. It then hooked up with Hadoop-specialists Hortonworks to bring out a Hadoop distro for both Azure and Windows Server.

"Microsoft recognizes Hadoop as a standard and is investing to ensure that it's an integral part of our enterprise offerings," the Redmond company wrote in a blog post on Monday.

HDInsight is based on the Hortonworks Data Platform and as yet is based on the original version of MapReduce, rather than the revamped YARN system found in the just-released Hadoop 2. Microsoft will support Hadoop v2 in a future update to HDInsight the company said.

Microsoft boasts of "deep integration" between HDInsight and Redmond tools such as PowerPivot, Power View and the Excel 2013 add-on Power Map.

Pricing for the service starts at a minimum of $0.32 per hour for a head node and $0.16 an hour for a compute node, rising to $0.64 an hour for a head node and $0.32 for a compute node from December 1 as the service exits preview pricing.

Other features, such as storage and data transfer, are billed as standard Azure pricing, which admins will need to watch given the elephant-logo'd tech's propensity to suck up as much data into its HDFS gut and keep it there. ®