Kyvos Insights, an OLAP-on-Hadoop startup, is announcing today that it can now run on Microsoft's Azure HDInsight. Customers can now provision an HDInsight cluster, then set up an Azure virtual machine, install Kyvos on it, and federate that Kyvos instance with the HDInsight cluster. With the announcement, the Kyvos/HDInsight combination is now both compatible and officially supported.
Welcome to the OLAP party
Kyvos joins Big Data vendors Platfora and AtScale in offering an OLAP cube approach to Hadoop analytics. There are differences though. Platfora doesn't call itself out as an OLAP solution -- instead it guides customers through the creation of "lenses," which run on Platfora's own in-memory platform, including dedicated hardware. AtScale, meanwhile, runs on Hadoop but, according to Kyvos, simply creates an OLAP semantic layer which, when queried, generates Hive or Impala queries on the fly.
Kyvos, on the other hand, configurably materializes its cubes, and does so in a distributed fashion, storing "cuboids" across cluster nodes in HDFS. Additionally, its engines for creating and querying the cubes run on the cluster as YARN processes.
The Redmondian force is strong in this one
Beyond this true OLAP aspect of its engineering, and the alignment with Azure HDInsight, Kyvos has an added OLAP/Microsoft angle to its product. Specifically, Kyvos is driver-compatible with Microsoft's OLAP platform, SQL Server Analysis Services (SSAS), and uses the MDX query language and XML for Analysis (XML/A) protocol from that platform.
This SSAS "emulation" (my term, not Kyvos') creates a high degree of compatibility for BI applications (including custom-developed ones) designed for SSAS. In effect, Kyvos would like you to think of its product as a scale-out, drop-in replacement for SSAS that requires little in the way of code modification or skill set changes. And its an OLAP platform that runs in the cloud -- something Microsoft offers only via Power BI, which has very real scale limitations.
For enterprises that are SSAS shops, Kyvos may have a very compelling offering here. It's important to realize though, that designing OLAP cubes is a process that is explicitly geared to a structured view of data. For some, that is the very antithesis of working with Hadoop. Others may not see it that way -- the idea of sticking with the OLAP paradigm while transcending its data volume limitations may turn out to be very pragmatic.
Disclosure: I work in a senior strategy role at Datameer, which is a competitor to Kyvos, and which promotes a schema-on-read/unstructured data paradigm for working with Hadoop. My briefing for this story was conducted by Kyvos Insights' VP of Product, Ajay Anand, who was Datameer's CEO when the company was founded. I made Mr. Anand aware of my role at Datameer before our briefing,