Question

All-

So I need to store 3D positions (x, y, z) associated with objects in a video game.

I’m curious, is this a terrible idea? The positions are generated quite frequently, and may vary some.

I basically would ONLY like to store the position in my database if it’s not within a yard of a position already stored.

I was basically selecting the existing positions for an object in the game (by object_id, object_type, continent and game_version), looping through, and calculating the distance using PHP. If It was > 1, I would insert it.

Now that i’m at about 7 million rows (obviously not for the same object), this isn’t efficient and the server I’m using is coming to a crawl.

Does anyone have any ideas on how I could better store this information? I’d prefer it be in MySQL somehow.

Here is the structure of the table:

object_id
object_type (like unit or game object)
x
y
z
continent (an object can be on more than one continent)
game_version (positions can vary based on the game version)

Later when I need to access the data, I basically only query it by object_id, object_type, continent, and game_version (so I have an index on these 4)

Thanks!
Josh

Answer 1

Presumably objects on different continents are considered infinitely far apart. Also you haven’t disclosed the units you’re using in your table. I’ll assume inches (of which there are 36 in a yard).

So, before you insert a point you need to determine whether you’re within a yard. To do this you’re going to need either the MySQL geo extension (which you can go read about) or separate indexes on at least your x and y columns, and maybe the z column.

Are there any points within a yard? This query will get you whether there are any points within the bounding box of +/- one yard around your new point. A ‘nearby’ result of one or more means you shouldn’t insert the new point.

  SELECT COUNT(*) nearby
    FROM table t
   WHERE t.x between (?xpos - 36) AND (?xpos + 36)
     AND t.y between (?ypos - 36) AND (?ypos + 36)
     AND t.z between (?zpos - 36) AND (?zpos + 36)
     AND t.continent = ?cpos

If you need the query to work with Cartesian distances rather than bounding boxes you can add a sum-of-squares distance computation. But I suspect bounding boxes will work just fine for your app, and be much more efficient than repeatedly fetching 75-row result sets to do proximity testing in your application.

Conceptually it wouldn’t be much harder to create a stored procedure for MySQL that would conditionally insert the new row only if it met the proximity criteria. That way you’d have a simple one-way transaction rather than server back-and-forth.