Optimal Location to Store for Interactive Retrieval in a Modern Data Web Application (and also ad-hoc reporting)

[#1]

Where to store a 672 dimension, million record dataset for online applications?

How would you store it, lay it out. Most queries are working with smaller subsets of the dimensions, say 20-30 at a time.

HBase or HBase + Phoenix has been thought of?

Or would Hive + Tez + ORC work well.

Should it be cached like Apache Ignite? Apache Geode? Redis?

Any suggestions? Looking for best practices for a greenfield application.

[#2]

Have you looked at Apache Kylin (which is built on top of hbase) ?

http://kylin.apache.org/

[#3]

See also

http://kylin.apache.org/docs15/gettingstarted/best_practices.html

[#4]

is Kylin a support part of the HDP 2.4 stack?

[#5 · Accepted solution]

Druid is another option and is supported in HDP 2.6