MgCooker: more than 90% of time wasted to build blank tiles

[znarf1965]

Dear Sirs,

Tile system is a great tool, fundamental, in some cases. The big problem is the time it takes to pre generate tiles. The performances in many cases are so bad that it is nearly impossible to generate tiles. The performances problem are mainly related to the huge amount of blank and useless tiles it creates.

I’ve made a very simple example regarding a very small town and very simple data. In this case and for scales 1:35000, 1:5000 and 1:2500 the generated tiles are more than 90% blank. If I try to tile closer scales (1:1000 or 1:200) the percentage get even worse.

I’ve made many tests on Mapguide 2.4 32bits (Windows 7) and 64 bits (Windows Server 2008) changing parameters and the best performances I get checking “Use official methods” and leaving “meters pr unit” = 1.

Nothing changes if I try to limit bounds by overriding. The parameters are not used and the estimation on the amount of tiles remains the same.

I've attached an mgp with a readme that explains my steps to generate the problem as I have on my environment.

Best regards

Francesco

[znarf1965]

Simple test project with a readme

[znarf1965]

Just a note to say that I've used Maestro 5.0b1

[znarf1965]

Sorry, I've changed priority to hight becasue some project really do not work without tiles.

[jng]

I think MgCooker is due for a rewrite. The tile-gen algorithm seems fundamentally flawed (or broke somewhere along the way) based on the numbers given.

Can you give me (if applicable):

a) The number of expected tiles to be generated (as reported by the tiler in AIMS) b) The ETA of that operation (as reported by the tiler in AIMS)

[jng]

Removed non-official tile generation method in r8118