arnulf.christl at CCGIS.DE
Fri Feb 18 14:07:22 EST 2005
> I'm sure you've all seen Google and its (obviously) pre-canned map
our story is exactly the same as Daniel told it - except we needed 3
years longer to understand what happened
> Personally I find that we've got a quite complete set of features, so my
> personal wishes would go towards improving what's there instead of
> adding new features:
And again, I can just second what Daniel said. The installation is OK
though, you already did a good job. What about security audits? That has
to be done in the architecture and clients as i understood it?! If you
need high security a java OWS Proxy would do the job. Why add this
uninteresting, bothersome and tedious domain to MapServer?
Docs and Demos:
The docs and demos need improvement - but hey, who wants to write
documents and prepare demo data?
That said we officially announce to provide for an easy to use and
extendable WMS demo with minimal data and minimal functionality (our 2
cents). Lately I had the chance to observe a beginner trying to get
itasca to run - it was rather painful.
There was a lot of talk on changing and improving docs. Has anything
come out of that?
Are google maps really that cool? After some toying around I dont think
so anymore. Again its just like Daniel said - they just discovered
technology that we have left behind years ago. They just throw disc
space and bandwidth at it and it works fine. But its static and when you
really count the time it needs to get all those tiles though the pipe to
your screen it alway takes longer than requesting one single map from an
otimized MapServer. Its the last mile - you never get the maps all at
once. To do a "benchmark" just count the seconds it takes to finish one
screen. Something is fidgeting all the time and makes me nervous, thats
all. All those tiles which are loaded on anticipatory obedience around
the visible map just produce traffic. People would hate me for a client
that did something like that. Its an extra 0f between 30 to 60% of what
you really need. Its all about pychology of perception.
Slightly blushing we must admit that we do exactly the same as google
maps and automatically rasterize vector data into tiles to increase
performance. It works great! After the initial job of tiling the whole
area (with MapServer) it gets slightly more sophisticated in that it
automatically generates tiles for those areas where vectors have been
modified. You only need a tag in the database each time an object is
changed. Every now and then (or even triggered by the database) the
bboxes of the tagged objects are intersected with the raster catalog and
whatever tile is affected is renewed. Done.
MapServer does this job perfectly, so I don't think anything much has to
be done inside it. It is rather the architecture around it and metadata
that can help. If this is interesting to anybody - it could be developed
into a sleek side project. The zoom levels have to be preset in the
client, MapServer shouldn't have to bother about that. Thinking about
how to synchronize static MS maps and overlayed dynamic maps with an OGC
client gives me headaches.
We need to do this (crap) only because official German land register
maps have to conform to a cartographic guideline from the mid 70s. To be
able to satisfy these guidelines we need to overlay 12 WMS with between
5 and 15 layers each. That means each single map request calls 12
separate MapServ threads and each opens 5 to 10 connect/read on a
PostgreSQL PostGIS box to render around 50 polygon styles, 200 line
styles and 900 symbol classes. Typical database counts are between 10
and 100 million objects. Updates by the hour. That really rocks.
98% of all maps are requested at scales between 1:500 and 1:2500. The
obvious thing to do is tile them. After that we still show them with
MapServer, as there is nothing much faster or more stable or better
We are very interested in further improving cartographic functionality.
Steve, Frank and Daniel saw some of it (and we have more for the next
conference to discuss :-). That is something where none of the US GIS
companies ever managed to live up to our standards. MapServer does!
(with an odd trick here and there... see the document by Peter Freimuth
If Cairo is a more powerful cartographic alternative couldn't it be
included as an option? Then GD would be fast and Cairo would be nice,
depending on what you need.
The new web site:
(where is it, will it be multilingual?). It would be extremly helpful if
we could continue further develoment discussions there in wiki style.
...and then again: MapServer is a very stable, very fast and very
professional software. Thanks to everybody involved!
Arnulf B. Christl
More information about the mapserver-dev