Re: Marker Point Display - Large data set

yeah, jon is right - as much as I like the system I use (the UK county
one you saw - http://gmaps.evilc.com ) it is very much up to what kind
of system you plan on implimenting.
Whilst designing my "manual bunching" system I have tried to acheive a
best-of-all-worlds solution for letting people browse a collection of
points in a meaningful heirarchical structure.
I put up some stuff on the mapki with some thoughts on the issue a
while back: http://mapki.com/index.php?title=Talk:Knowledge_Base but it
is quite out of date. I have acheived all of the things I was pondering
and the SQL has changed a bit since then but as I have said to people
before, I am quite busy but if someone wants to use the system I would
be willing to take the time to document the techniques properly.
It uses SQL to decide what should show and what shouldn't given any
centre and zoom. Pulling all visible points is done in one SQL
statement (using nested set joins to walk the heirarchy trees) so it
should be efficient and portable to any language. (eg PHP, Python etc).
It should also be compatible with any flavour of SQL.

I have also been thinking lately that my system could also be combined
with other bandwidth / plot time saving techniques such as only loading
the first X points then loading X more when more is clicked.
However, I do wish to avoid that - it just seems too hit & miss to me.

I would imagine that any system that, like mine, only loads points in
or near the viewport (bounds) should theoretically be able to support
the same number of points / m2 at zoomlevel 0.
Lets say you worked out the # of points plottable in one go before it
gets too slow and came up with around 1000.
Then your max point density would therefore be whatever 1000 divided by
the number of metres square you see at zoomlevel 0 is or some such
calculation like that. It would probably be dependant on the size of
the map window too.

Whatever, that is only at zoom 0. Above that, realistically, you are
probably going to want some kind of "bunching" system to reduce that
down.
If that can be done, then you should be able to deal with anything as
long as you are smart with the bunching so as not to ever be able to
see more than the 1000 (or whatever it is) points.
Without bunching, then you would have to limit the points loaded and
have a "plot more..." button, or just make sure you never went above
1000 (or whatever it is) points, unless someone has another idea.

I would be extremely interested to see some benchmarks done on the
various techniques.
As the US county / state border data is publicly available and free, I
was thinking of trying to write something that would randomly add
points in the US to my system and also set the region id to the right
one for any given point.
Then see at what point my system falls over. I don't see why I
shouldn't be able to hit a million points as long as they are spread
out so that you didn't go over the 1000 point (or whatever) figure for
any one view. Now whether it's a million points worldwide, or a million
in the US is another matter, but I would still like to see six figures
in the mouseover on my site that shows how many points are in that
region point :)