I was recently fortunate to speak at one of the database technology conferences held in Israel. Big parts of this conference revolved around ‘Big Data’ and I was asked to give the Business Intelligence perspective on this fascinating subject.
columnar database technology on the basic premise of business intelligence - the ability to have business users perform ad-hoc analytics and reporting tasks over as much data as possible.
In order to do that, I represented a business user building a report over a very large operational database containing 13 tables, the largest of which hold 100 million and 40 million rows. While databases of this size were once rare - now, any company who has a properly tracked website quickly accumulates even more data than that.
To demonstrate, I was using a front-end analytics tool (SiSense Prism) to create reports that query the database directly - a feat not advisable with a relational database. So instead of querying the source database, the data was replicated (but unmodified) into a columnar database that was designed specifically for ad-hoc analytics - ElastiCube.
The computer holding the ElastiCube was a $1200 off-the-shelf PC with 6GB of RAM, 100GB of disk space and a single quad-core CPU (64-bit). The Prism front end could be installed on any computer, as it does not process the queries or hold data - only requests query results.
For your convenience, here is the video of this demonstration.
One thing you should understand from this video is how simple drag-and-drop operations of a business user (or multiple users) within his or her desktop tool turn into complex database operations that would choke any relational database (joining, grouping, aggregating), yet are handled by a columnar database without any difficulty.
Which brings me to my final, and perhaps most important point -
Whichever business intelligence front end tool you pick, dealing with issues like this (and their subsequent side effects) is in fact 90% of the life-time cost of a BI solution, and often why the solution stops being used all together. This is because BI solutions which rely on relational back-end technology must be designed to assume there had been significant trimming, de-normalizing and pre-aggregation of the data conducted prior to being delivered to business users. This process never ends, and only becomes more and more difficult to maintain over time.
Columnar databases change this reality entirely, and combining them with 64-bit and multi-core computing makes for a dramatic evolution in BI development.
By: Elad Israeli | The ElastiCube Chronicles - Business Intelligence Blog