So those older ways of defining metrics and dimensions for use, like OLAP cubes, were difficult to maintain and inflexible. They became stale on arrival, the business needs usually had moved on by the time development had been completed.
Newer systems like MetricFlow, Cube and AtScale are much better. They can define a whole data model and allow access at any grain.
I really like to play around with the different implementations so far. But it is still a lot of playing around. I have a bit of a LookML past with all the pros/cons (you covered them well).
From the integrated ones I enjoy the Lightdash one most - it's already where I usualy prepare the data (dbt), their approach makes more sense than the dbt one (at least for me). And like you, I am hoping there will be more people asking for charts and dashboards as code.
In two new projects I am starting to work with Cube in small scale scenario and so far I really like to work with it. But I am still scratching the surface.
Our managed metrics need this flexibility. Especially on the time dimension but also on any hierarchical dimension. This will be an important layer for us.
I also especially like Cube’s multi-API approach. They are investing in the SQL API, and we have run some experiments with positive results of connecting various BI tools to it.
The reuse benefit of pushing metrics and semantics into the warehouse is huge. That allows our managed metrics to be consumed by virtually any product.
Hey David, good stuff
You have a great series of articles about the semantic layer; I'm learning a lot from you.
i have one question, tho.
what you meant with
> I hope we are in agreement that things like OLAP cubes in their original implementations (like Microsoft SSAS) aren’t a good idea going forwards
Thanks Thiago!
So those older ways of defining metrics and dimensions for use, like OLAP cubes, were difficult to maintain and inflexible. They became stale on arrival, the business needs usually had moved on by the time development had been completed.
Newer systems like MetricFlow, Cube and AtScale are much better. They can define a whole data model and allow access at any grain.
I really like to play around with the different implementations so far. But it is still a lot of playing around. I have a bit of a LookML past with all the pros/cons (you covered them well).
From the integrated ones I enjoy the Lightdash one most - it's already where I usualy prepare the data (dbt), their approach makes more sense than the dbt one (at least for me). And like you, I am hoping there will be more people asking for charts and dashboards as code.
In two new projects I am starting to work with Cube in small scale scenario and so far I really like to work with it. But I am still scratching the surface.
Yeah we've enjoyed working with Cube too, their API is really good.
This is helpful, David. The semantics layer is in flux and this article helped me understand what’s going on at the moment
Glad to hear it! What's your plan for using one in Datateer?
Our managed metrics need this flexibility. Especially on the time dimension but also on any hierarchical dimension. This will be an important layer for us.
I also especially like Cube’s multi-API approach. They are investing in the SQL API, and we have run some experiments with positive results of connecting various BI tools to it.
The reuse benefit of pushing metrics and semantics into the warehouse is huge. That allows our managed metrics to be consumed by virtually any product.
GoodData is a BI tool with an open semantic layer - best of both worlds!?
I'm going to take a look at GoodData soon hopefully