Normalisation vs Performace: benefit/issues of removing linking tables in (this) schema?

Question

Generally i like to keep my database as clean and expandable as possible.

However after doing some tests, I realised that whilst this is usually the best way to do it, when dealing with large datasets its a lot slower than what i refer to as the "dirty" approach to the problem.

Basically lets say i have a table of objects. These objects belong to certain people. One object may have one person, whilst others more than 1. My initial thought was as I always do, create an objects table for my objects, a peoples table for my people, and then a object_to_people linker table.

However joining the object and linker table in order to get all objects a person is assigned to, can take up to 3 seconds (that's based on around 400k records, but only 1 link per object). Yes i also set up index's e.c.t. to try and speed things up.

If I instead remove the people and linker table, and put the people in the objects table as columns and use 1/0 to set whether each person is assigned to that object, without joining the two large tables i see a speed of around 0.3 -> 0.7 seconds (varied greatly).

To begin with, we only need 2 people. But I don't want to be too restrictive if i can help it. I know I can use caching and what not to improve the end user timings, but is there any reason this would be considered a really bad idea to use columns rather than link tables?

Answer 1

I have a similar setup.
My join table has 17,000,000 rows. My "person" table has 8,400,000 rows, and my "objects" table has 300,000 rows.

I have queries with multiple joins on my joins table and unions of results that return tens of thousands of rows and they take less than 1 second to run (50-400ms).

I think your first layout could be fine, but you probably need to focus on your indexes and queries.