James Dixon Imagines A Data Lake That Matters


Dan Woods, Contributor

January 23, 2015

While much has been said and written about the data lake, in the end, the concept has stayed pretty simple. The idea has been to use Hadoop as a place that data of all types could be stored in greater detail than ever before, at an affordable cost, and then used to power the existing data warehouse ecosystem and also perform new types of analytics.

This week, James Dixon, the CTO of Pentaho and the creator of the term “data lake”, presents a challenge to the big data community in his latest blog “Union of the State – A Data Lake Use Case”. Dixon argues that it is time to start figuring out how to make the data lake a time machine for a business.

Here is his idea in a nutshell:

  • Most of the applications in use capture the state of the enterprise in some way.
  • A fraction of the changes to the state are captured via change logs.
  • Dixon suggests to store all of enterprise data in a data lake and effectively create change logs for every field.
  • In addition, Dixon suggests to capture behavioral data for how applications of all types were used.
  • in this way the data lake comes a time machine that allows the state of the enterprise at any one moment to be captured and analyzed.
  • The power to see what happened before and after important events will lead to even better predictive models and a deeper understanding of operations.

What I like about Dixon’s visions is that he gives the data lake an ambitious mission. But the how is important. In his blog Dixon addresses a few implementation details. I want to go further here add some suggestions for how to make the Union of the State vision a reality.

Building a Union of the State Machine

Dixon’s vision is that by either using change logs, databases logs, or taking a snapshot of the state of a database, you will be able to create the time machine for enterprise data of sufficient granularity to matter.

When I started thinking about Dixon’s vision it occurred to me that there must be a way to make creating the Union of the State machine, one that has full history, by using some existing technology. Instead of a naive implementation in which you derived change logs from snapshots using direct comparisons, why not use databases that already have timestamps embedded in them. Teradata’s Integrated Data Warehouse has such a capability. So does Datomic, a so-called immutable database that preserves all state ever entered. Using one of these two approaches could make building a Union of the State machine much easier.

The second way these technologies would help would be in running queries. Both Teradata and Datomic have query languages that allow you to set the point in time for the state you want to consider and discover. I’d be interested in finding out about other repositories that have the same ability to handle a time stamp. Splunk could do the time stamp just fine, for example, but I suspect would have a harder time with the aggregations, although I’m sure a Splunk master could manage it.

The next challenge would be creating the predictive models that looked at the way the data changed over time. Nutonian could play a hugely positive role here to automatically create models. All of the statistical modeling tools such as R and equivalents could also play a role.

Of course, you would have to focus. Right now, most data lakes are operationalized through some sort of SQL that contains a fraction of data in the data lake that has been distilled. Teradata has an excellent way to talk about this data. They say it has high business value density. It is unlikely that even the largest companies would have the the ability create a complete Union of the State machine that included all data from all apps. Getting started by focusing on data that has the highest business value density will be a necessity.

While this outline of an implementation plan is far from exhaustive, it does show how many existing technologies could play a role at bringing Dixon’s vision to life. If you have more suggestions for other technology that could help, please drop me a line.

Follow Dan Woods on Twitter:

Follow @danwoodsearly

Dan Woods is on a mission to help people find the technology they need to succeed. Users of technology should visit CITO Research, a publication where early adopters find technology that matters. Vendors should visit Evolved Media for advice about how to find the right buyers. See list of Dan’s clients on this page.

This article was written by Dan Woods from Forbes and was legally licensed through the NewsCred publisher network.

Great ! Thanks for your subscription !

You will soon receive the first Content Loop Newsletter