On Tue, Nov 3, 2009 at 9:05 AM, Josh Livni <[email protected]> wrote:

> With that side rant over, I also think that this mindset implies that one
> global layer to rule them all is the best way to go:  After all, we only
> care about processed tiles, so lets just make sure every tile is the "best"
> available for the area in question, and we're done.  I don't see it that
> way.

Nor I. My proposal probably doesn't surface this enough (you'd have to
read down into the DHT hash key proposal) but I am assuming

(a) we will have multiple "layers"
(b) for each layer there will be one "canonical view", the "best"
view, with a fixed known address for each tile
(c) for each layer there will be one "canonical metadata address", an
address to get a metadata block about that tile
(d) inside that metadata block, will be information about all the
*other* information in that tile beyond the current canonical view
(e) so, for example, when new imagery supersedes old imagery, the old
imagery doesn't go away, it continues to exist in a non-canonical
address for that tile, and is findable through the standard metadata
address for that tile
(f) so, the data update process is one where a new tile is first
written into a unique address, where it will persist "forever" and
also written into the canonical address, where it will reside until it
too is superseded

This view is still tile centric and doesn't make any allowance for
storing the raw imagery, though probably we could find a place in the
addressing scheme to chop the original unprojected imagery into
storable chunks and dump them into the DHT too, just not as objects
useful for mapping.

P.

_______________________________________________
talk mailing list
[email protected]
http://openaerialmap.org/mailman/listinfo/talk_openaerialmap.org

Reply via email to