[OSM-dev] Extra-large metatiles

[Christoph Hormann]

On Monday 15 January 2018, Paul Norman wrote:
>
> Latency does not matter with a pre-render task, so I've been
> considering using extra-large metatiles. Has anyone tried this
> before?

Not for normal rendering but for raster based data 
pre-processing/generalization tasks i am usually working with 16k 
metatiles.  There are mostly two things that limit the usefulness of 
increasing tile size in this field - which might to some extent also 
apply to normal rendering:

* memory footprint:  Since the processing itself is usually single 
threaded you run several tiles in parallel and the memory demands for 
this must stay within the available memory.
* processing time granularity:  Since different parts of Earth take very 
different times to process it can happen that the slowest processing 
tiles are processed last leading to less than optimal parallelity in 
processing overall (i.e. you have one or two tiles dragging on in the 
end using only part of the ressources available while the rest is 
already finished).

> Based on experience from metatiles 1-8 tiles across, as size
> increases I expect query time and time to render to increase, but
> time per tile to decrease. I don't know at what point this stops
> being true.

I would expect this to happen pretty fast, possibly 8x8 is already at or 
near the top.  But it will likely depend quite a bit on the hardware 
used and the specific rendering task.

-- 
Christoph Hormann
http://www.imagico.de/