Press "Enter" to skip to content

Day: November 22, 2019

Columnar File Formats in Hadoop

Matthew Rathbone gives us an overview of the benefits behind the ORC and Parquet file formats:

People throw this term around a lot, but I don’t think it is always clear exactly what this means in practice.

The textbook definition is that columnar file formats store data by column, not by row. CSV, TSV, JSON, and Avro, are traditional row-based file formats. Parquet, and ORC file are columnar file formats.

Read on for a comparison and example. In the SQL Server world, think columnstore versus rowstore indexes and you won’t be too far off.

Comments closed

Aggregations in Power BI

Shabnam Watson takes us through aggregations in Power BI:

In Power BI, Aggregations start as tables just like any other table in a model. They can be based off a view or table in the source database, or created in Power BI with Power Query. They can be in Import or Direct Query storage mode.

Once in the model, these tables can be configured so that the engine can use them instead of a detail table to answer queries when possible. The process of creating and configuring aggregations in Power BI is significantly easier than the process of creating aggregations in SSAS multidimensional.

Once an aggregation table is configured, it becomes hidden from end users. Report developers and end users don’t know that it exists and don’t need to change anything in how they query the dataset.

This was one of the key benefits to a multidimensional model. Shabnam has an excellent, detailed article here, so give it a read if you are a Power BI developer.

Comments closed

Decomposition Trees in Power BI

Tomaz Kastrun takes us through a new visual in Power BI:

Decomposition tree is a data presentation of slicing and dicing of selected metrics based on the attributes of these metrics or with combination of other metrics. Another great aspect of this visual is to analyze the selected variable with many metrics or attributes (dimensions) as the same time.

It’s not the type of visual I’d want to see on a dashboard, but I can see it as quite useful in exploratory data analysis.

Comments closed