DATA MESH AND DATA WAREHOUSE

DATA MESH AND DATA WAREHOUSE:

BUILDING ON A FOUNDATION

By W H Inmon

Every new generation in IT has a fundamental choice to make – do I build on the existing foundation or do I tear it down? Once upon a time it made sense to tear it down. When the computer first appeared using the automation afforded by the computer, the computer paved the way for tearing down things that were done manually. That destruction was a very constructive form of destruction.

But is tearing something down something that already exists always a good thing to do? The people at data mesh have inferred that with data mesh you don’t need a data warehouse. The people that promote data mesh think that at last we are freed from the tyranny of having to build a data warehouse.

Let’s examine this line of thought.

What is the benefit of building a data warehouse? What is it that a data warehouse brings to you? A data warehouse brings you believable, vetted data. Once you have built the data warehouse properly you can trust the data that you have put into the data warehouse. The real benefit of the data warehouse – if built properly – is believability of data, having confidence in what the data represents.

So what happens when you build the data mesh on top of data that is not believable? You get an eternal uneasiness about the data. You never really know whether your data you are operating on is really believable. But if you build your data mesh on top of a data warehouse, you have confidence in your data.

The question then really boils down to – do you want to have confidence in your results or not? If you don’t care about having confidence in your results then you don’t need a data warehouse when you build your data mesh. But if you care about having confidence in your results you need a data warehouse. It is as simple as that.

Data mesh is not a replacement or substitute for a data warehouse.

So what do you gain by having data mesh without having a data warehouse? If you go that direction, you don’t have to go through the awful and feared exercise of having to integrate your data. Every consultant and every vendor in IT runs from integration like Satan himself was coming with his pitchfork. Consultants and vendors absolutely despise having to integrate data.

But the fact is that the price of having confidence in your data MANDATES that the data be integrated. You CANNOT have confidence in your data unless your data is integrated. This simple fact is non negotiable.

If data mesh were to offer itself as an extension of believability, then great things will happen. But as long as data mesh offers itself as an alternative to operating on believable data, then data mesh is doomed to go the way of other failures in our IT profession.

Building data mesh on top of an unvetted, unbelievable foundation is like the skyscraper in San Francisco that was not built on bedrock. You don’t want to be on Market Street or Chinatown when the building comes crashing down. Nothing good is going to happen then.

No alt text provided for this image

But this principle is not limited to data mesh. This is true of a whole bevy of other technologies that depend on operating on believable data. This principle is true for AI, ML, OLAP, the dimensional model, and a whole host of other technologies.

Once upon a time it may have been useful to destroy the existing foundation. But that day is no longer. It is time to build on the existing foundation.

Bill Inmon lives in Denver, Colorado with his wife and his Scotty dogs Jeb and Lena. Lena is younger than Jeb and runs faster than he can. But Jeb still can run fast for his dinner.

Marcin Szpot

Senior Cloud Data Engineer at Future Processing

2 年

My two cents (as a data engineer). From readings I did my understanding is that DWH and Data Mesh can co-exist simultaneously. Key element that points 'which path to choose' is a question 'where this data will be used'. DWH will remain as a key analytics/kpi/clean metrics solution since its capabilities to store and aggregate data. Data Mesh on the other hand could be used for robust/on-demand/real-time analytics where those who inject data into it are held accountable for quality of that data; this leads to not only tech stack change but also change in co-workers trust. Coming of Data Mesh could benefit both worlds by bringing division and unloading tasks that dwh simply can't handle/is not designed for. Only risk of this co-existence is danger of unecessery workforce. Data Mesh is a great opportunity for more technical staff to be nominated as a 'tech data steward' of their department.

回复
Sanjeev Bhattacharya

Advisory Services | GCC Leadership | Digital Transformation | Data Privacy & Security | Advanced Analytics | AI/ML

2 年

Correct! There is no shortcut to success.

回复
Alessandro Crapisi

EMEA Software & Data Architect Lead

2 年

Agree ! And I think data confidence comes only after deeper data checks with SME

回复
Nick Stachniak

Solutions Architect, ArangoDB

2 年

Love it!

回复
Vijay Yadav

Head of Data Science ? AI/ML Leadership ? Digital Transformation Strategist ? P&L Impact in Pharmaceuticals & Life Sciences ? Advanced Analytics & Predictive Modeling ? Data Governance ? LLM & Generative AI Innovation

2 年

Bill Inmon Great article and you have simplied the thinking really well! I love it ! I respectfully have different point of view on couple of things: 1. The assumption that data in DW is always believable, is not the case in real world. Ideally that is how it should be. Readers should not assume by default that data in DW is always believeable. They need to verify and take remedial action to make it believable. I thought that disclaimer is important in your article. The lineage, quality got to be ensured as well. 2. Integration means different things to different people. In order to mesh to work in real world, data should be intergated/connected at the business domain and business process levels. This is important because if there is problem in downstream business process, we should be able to lineage back to upstream process. Unfortunately, This is not the case in real world. In order for Data mesh to work, all business domains should be integrated in addition to business process integration. Point here is that integration is needed for Data Mesh to work.

回复

要查看或添加评论,请登录

Bill Inmon的更多文章

  • STREAMLINING THE EMERGENCY ROOM - TEXTUAL ETL

    STREAMLINING THE EMERGENCY ROOM - TEXTUAL ETL

    STREAMLINING THE EMERGENCY ROOM By W H Inmon The emergency room of the hospital is where people turn to when they have…

    2 条评论
  • THE TEXT MAZE

    THE TEXT MAZE

    THE TEXT MAZE By W H Inmon A really interesting question is – why does text befuddle the computer? The fact that 80% or…

    2 条评论
  • BLAME IT ALL ON GRACE HOPPER

    BLAME IT ALL ON GRACE HOPPER

    BLAME IT ALL ON GRACE HOPPER By W H Inmon One of the more interesting aspects about the world of IT is that IT people…

    17 条评论
  • ASSOCIATIVE RECALL AND REALITY

    ASSOCIATIVE RECALL AND REALITY

    ASSOCIATIVE RECALL AND REALITY By W H Inmon A while back, on a Saturday night, my wife and I were looking for a movie…

    7 条评论
  • A FIRESIDE CHAT WITH BILL INMON

    A FIRESIDE CHAT WITH BILL INMON

    A FIRESIDE CHAT WITH BILL INMON Get Bill’s perspective on your IT organization and its initiatives. Come spend an hour…

  • MESSAGE TO ELON

    MESSAGE TO ELON

    MESSAGE TO ELON By W H Inmon Yesterday Elon Musk tweeted a message asking if anyone had some innovative ways to improve…

    73 条评论
  • GREAT EXPECTATIONS:WALT DISNEY AND THE PENTAGON

    GREAT EXPECTATIONS:WALT DISNEY AND THE PENTAGON

    GREAT EXPECTATIONS: WALT DISNEY AND THE PENTAGON By W H Inmon Think of all the delight Walt Disney has brought the…

    5 条评论
  • BUILDING THE LLM - PART VI

    BUILDING THE LLM - PART VI

    BUILDING THE LLM – Part VI By W H Inmon The language model is an interesting piece of technology. There are many facets…

    3 条评论
  • BUILDING THE LLM - PART V

    BUILDING THE LLM - PART V

    BUILDING THE LLM – Part V By W H Inmon The generic industry language model has at a minimum three important elements of…

    2 条评论
  • BUILDING THE LLM - PART IV

    BUILDING THE LLM - PART IV

    BUILDING THE LLM – Part IV By W H Inmon The value of a generic industry language model becomes apparent when looking at…

    2 条评论

社区洞察

其他会员也浏览了