News and events

Kognitio 8.2 delivers new intelligent parallelism

New release automatically and dynamically deploys the platform’s processing power to provide the optimum combination of individual query performance and concurrent query throughput

UK, Bracknell June 19, 2017 — Today Kognitio announced the general availability of Kognitio Analytical Platform 8.2. The new release is focused around improving the way Kognitio handles very large levels of concurrency, allowing it to work well over a wide range of query complexity and concurrency levels.

Kognitio’s new intelligent parallelism feature allows the software to choose different levels of parallelism for different parts of a query at runtime, so large steps run on all available compute nodes, but smaller steps run on fewer. The level of parallelism scales automatically at runtime depending on the overall load on the system. So as more jobs are run at once, each job is broken into a smaller number of pieces.

Andrew MacLean, Kognitio Chief Technology Officer, explains, “If you have a hundred nodes then breaking one operation into 100 pieces makes sense, but breaking 500 operations into 100 pieces is 50,000 pieces. This makes a lot less sense because each node is scheduling 500 different operations which have a setup and teardown cost, and which need to be organized, talk amongst each other, etc.”

These enhancements in Kognitio 8.2 scored double the throughput of the previous release in Kognitio’s heavy concurrency benchmarks and allowed one of its workloads to be restructured to run with 1/20th of the memory footprint.

In addition 8.2 includes:

  • Offloading metadata and system management queries to a special, dedicated set of processes, ensuring good performance for system management and SQL compilation operations, regardless of system load.
  • Support for loading data in the JSON format
  • Support for authentication using Kerberos and Active Directory
  • An extensive set of OGC standard geospatial functions

“Having Hadoop-based data that is only accessible to a few expensive data scientists is no longer acceptable to organizations trying to get an ROI from their Hadoop adoption” said Roger Gaskell. Kognitio CEO, “Modern visualization tools such as Tableau allow business users to self-serve, but require an SQL interface into Hadoop that is fast enough to support interactive access at the same time as being able to support large user groups”.