Let’s cut right to the chase: you're reading this blog because you want expertise on storing data for your current data-driven business needs. Well, you've come to the right place!
Throughout this post, we'll look at the definitions and sample use cases for data vaults, warehouses, lakes and hubs. The differences between them are subtle, but they all serve a different purpose in the data world today.
A data vault is a system made up of a model, methodology and architecture that is specifically designed to solve a complete business problem as requirements change. So, as your business requirements morph over time, the data vault will maintain the historical system of reference or archive of your data and easily relate it to the new standard of data that you have defined. I like to think of the data vault as a customized, dynamic solution that gives business users access to all data (current and historical).
The biggest data vault use case is when a business, such a bank, needs to audit their data.
Let’s say you decide you need to update your security model to include additional fields and new applications in your enterprise. Using a data vault, you are able to checkpoint the time you made the security model changes and update your infrastructure with the changes, including all associated applications. This means the business team continues receiving the full view of historical and current information regarding the audit trail.
A data warehouse is a consolidated, structured repository for storing data assets. Data warehouses will store data in one of two ways: Star Schema or 3NF, but these are only fundamental principles in how you store your data model. We have seen, advised, and implemented both principles, but the one major flaw is that everything must be strictly defined (both in schema and integration).
The most common use case for creating and using a data warehouse is to consolidate data and answer a business-related question. This question may be, 'How many users are visiting my product pages from North America?' This ties together the information you're receiving from your end users with a business question that needs to be answered from a structured data set. This is what most would identify as the cookie cutter business intelligence solution.
Read more: Data Warehousing with CloverDX
But, there is an alternative approach that is becoming more popular, especially when you are talking about cloud and more powerful warehouses.
Organizations are adopting the ELT approach. This entails “staging” their data in their warehouse (such as HP Vertica), and then letting the power of the database perform the traditional transformation. Essentially, you are performing the most expensive operations with a system where you have more resources.
How to get your data to your cloud data warehouse? There are several options. This clip is from our webinar on Data Ingestion into S3, Azure Blob, Redshift, Snowflake: What Are Your Options?
A data lake is a term that represents a methodology of storing raw data in a single repository. The type of data that’s stored in the lake does not matter and could be unstructured, structured, semi-structured, or binary. The fundamental idea for a data lake is to make available any/all data from applications so your data team can provide insights on a business problem or value proposition.
But the challenge begins when you want to try to make sense of your data. If you are dumping data into a data lake, how do you know what data you need and what data you don’t need? How do you determine where the data resides in the lake? This very quickly can become a data swamp if not managed correctly.
The use cases we see for creating a data lake revolve around reporting, visualization, analytics, and machine learning.
Here is the architecture we see evolving:
A data hub is a centralized system where data is stored, defined, and served from. We like to think of it as a hybrid of a data lake and a database warehouse, as it provides a central repository for your applications to dump data. It also adds a level of harmonization at ingest so the data is indexed and can easily be queried.
Please note that this is not the same as a data warehouse architecture, as the ETL processing is merely for indexing the data you have rather than mapping it into a strict structure. The challenge comes when you have to implement the data hub and how can you harmonize all of your siloed data sources.
In general, we see the same use cases for a data hub as we would for a data lake: reporting, visualization, analytics, and machine learning.
Hopefully, you have learned a little bit about each of these data models, as well as their individual values in dealing with multi-structured data.
At the end of the day, there is not one model or technology that's superior to the other. It varies for each use case.
This means that you must analyze your requirements, needs, and budget before deciding which approach to use. Technology is constantly evolving, and each of these models will evolve with it.
Discover more: Data Architecture