- Attacking a Big Data project with an all-or-nothing mindset leads to failure
- Breaking the project into manageable phases called crawl, walk, and run is recommended
- Crawling phase involves doing the minimum to start using Big Data, such as getting current data into Hadoop and setting up systems to bring in data
- ETL coding and data normalization are done in this phase to set up for success in analyzing the data
- Walking phase involves building on the foundation built in the crawling phase and starting to gain value from the data
Crawl, Walk, Run with Big Data
Attacking a Big Data project with an all-or-nothing mindset leads to an absolute failure. I highly suggest breaking the overall project into more manageable phases. These phases are called crawl, walk, and run.
In this phase, you’re doing the absolute minimum to start using Big Data. This might be as a simple as getting your current data into Hadoop. You’ll also set up the systems to continually bring in data.
In this phase you’ll start on your ETL coding and data normalization. This will set you up for success as you start analyzing the data. This phase has minimal amounts of analysis. Your focus is on creating the system that will make it as simple as possible to analyze the data.
In this phase, you’re building on the solid foundation you built while crawling. Everything is ready for you to start gaining value from your data.
In this phase you’re starting to really analyze your data. You’re using the best tool for the analyses because the cluster is already running the right tools.
You’re creating data products. These aren’t just entries in a database table; these are in-depth analysis of business value. At this point, you should be creating direct and quantifiable business value.
You’re starting to look at data science and machine learning as ways to improve your analysis or data products.
In this phase, you’re moving into the advanced parts of Big Data architectures. You’re gaining the maximum amount of value from your data.
You’re also looking at how batch-oriented systems are holding you back. You start looking at real-time systems.
You’re looking at how to optimize your storage and retrieval of data. This might be choosing a better storage format or working with a NoSQL database.
You’re using machine learning and data science to their fullest potential.
If you go directly from crawling to running, you’ll trip and fall. You’ll continually trip and fall and not know why. This is a common issue with new data engineering teams. They don’t set up a good foundation before they more on to the advanced phases.