[mapserver-users] Re: FW: [Manifold] Map Server Docs

[Dimitri Rotow]

>
> Dimitri Rotow wrote:
> >
> > Although this is a Manifold list and not a general GIS list, I
> hope people
> > will forgive me for going "off topic" slightly and commenting on the
> > Minnesota server.  It has pluses and minuses: the biggest
> pluses are that it
> > is free and runs on Linux.  The biggest minuses are that is is
> complex to
> > configure and operate, it is a low performance architecture and
> that it has
> > numerous security holes, the most obvious of which is that it requires
> > giving the Internet client "write" permission to the server's Internet
> > directories.  These make it risky to deploy in production environments.
> >
>

Daniel,

As much as I enjoy discussing the pros and cons of various servers, I must
gently object to your selecting only a portion of my original posting to
Manifold-L for cross posting elsewhere.  I apologize in advance to any
readers of other lists or the cc: list who have no interest in this
interchange... if so, please flame Daniel for cross posting and not me! :-)

You clipped out the main content of my posting.  My original posting went on
to say that I felt the main issue with the Minnesota server is that in
practise it assumes one also has a GIS for the nuts and bolts of mapping,
but that the details of configuring and operating the Minnesota server are
very different from any existing GIS.  That requires the operator to become
proficient in two worlds: the GIS and the map server.

I then pointed out the advantages accruing when an extensive GIS also
incorporates a map server: people need become proficient in only one system
and the presentation (publishing via Internet) can be as simple as a
one-click process.  This was the main point of my posting that touched on
the Minnesota server.

> I agree with some of your points, but I strongly disagree on the "low
> performance and numerous security holes"... you also forgot to mention
> (in the pluses) reliability/stability and connectivity with other

I agree that my phrase would have been better to write "the biggest pluses
are that it is free, comprehensive and runs on Linux."   However, since my
posting was to a peer group who (presumably) all already have Manifold and
thus Manifold IMS (Internet Map Server), I was not writing as carefully as I
would have for a general audience.  I was providing a few words so people
who have not heard of the Minnesota server would know what I was talking
about.  Plus, my point was not to write a technical review, it was to state
that I felt it was a better approach for a full-featured GIS to include a
map server than to have a disjoint toolkit of a GIS and a map server that
are fundamentally different in their operating styles.

> systems via protocols such as OGC WMS (client and server), and
> on-the-fly access to various data sources such as SDE, PostGIS, 20
> raster formats and half a dozen vector formats.
>

OGC and the like are fairly typical of the frightfully bureaucratized
approaches one sees coming out of the legacy GIS world.  Lots of raster and
vector formats are fine, but frankly, most of the nearly 200 million people
in the world who operate Microsoft Office don't care about OGC or SDE or
PostGIS. Those are the people who buy large volume and, ultimately, decide
what is widespread in the market.

I see parallels to how UNIX marginalized itself. Rather than pay attention
to what hundreds of millions of people felt was important, the UNIX
community wasted time and energy in elitist "standards" wars and standards
"processes" that no one except a fraction of a percent of computer users
cared about.  In the 1980's I knew many people in UNIX who felt that
adoption by the Federal government of certain UNIX-driven standards would
"force" the masses to accept what we were doing.  It was almost as though if
people could somehow cram a sufficiently encompassing synthesis of Berkeley,
AT&T System III and System V, and SUN approaches to file systems,
networking, etc., into one UNIX it would be accepted by the masses.  Of
course, the masses didn't care.

What I think is especially galling to UNIX/Linux partisans is that the
reason Microsoft has triumphed is that users have *chosen* Microsoft over
UNIX/Linux.  I have heard this explained away by fanatics using endless
rationalizations (Microsoft forces people to buy its products, hundreds of
millions of people are too stupid to know they should be buying what I
prefer, etc., etc.), but always avoiding confronting the key observation
that by margins of about a hundred to one [counting all the ripped-off
copies of MS stuff] people voluntarily prefer Microsoft to Linux.

I don't disagree that quite a few people prefer OGC and actually intend to
buy SDE and deploy it, think about PostGIS and so on. But they are in the 1%
cohort compared to those who don't care about such things and prefer a
one-click map server that allows them to present the data they have over
Internet.   One might say the unwashed 99% are too stupid to realize that
they should take OGC seriously, that they don't know what they want, that
they should be using Linux, and so on.  But in the end if one ignores the
legitimate interests of millions of people one ends up being marginalized.
Don't make the mistake of UNIX and assume away the "ease of use" thing, or
allow a dislike of Microsoft to not consider the desiderata of the only
(numerically) market that counts.

> I'm actually glad to hear that you have been able to combine easy
> configuration and application development, flexibility, high performance
> and affordable price into a single product... it's a huge step in the
> right direction since none of the current map server offering could do
> *all* of that until now AFAIK... not even MapServer which I have to
> admit requires some skills and some time before you can take full
> advantage of all its features.   This is not sarcasm... if what you
> pretend is true then I'm impressed!  But I won't believe until I see
    ^^^

Perhaps this is a simple case of the impression in writing being different
than what was intended, but I wish you would not have used quite so an
implication-laden word as "pretend."

Daniel, considering that I was writing in a list where all of us have
Manifold IMS there is no need to "pretend."  It's what we use on a daily
basis.  Manifold 5.00 does indeed cost $245, it does indeed (in addition to
what may be the most extensive and powerful general purpose GIS ever
introduced) include Manifold IMS at no additional charge, it does provide
superb performance in both GIS and IMS and one can, in fact, publish a map
server web page in it with one click [the "OK" button in the File - Export -
Web Page dialog using default settings].

A web page for t5.00 is at http://www.manifold.net/news/news_set.html,
including links to some of the beta sites for Manifold IMS and links to the
User Manual at http://www.manifold.net/manuals/5_userman/start.htm ... The
user manual lays out everything very openly (albeit in annoying format given
the automatic translator to HTML that was used).

> it...
>
> Does your note imply that Manifold IMS will deliver better performance
> (and stability) than MapServer?  Other than the mythic belief that CGI

That remains to be seen and quite likely (like many things in computing)
will depend in part on the specific application. Like anything, Manifold IMS
and MapServer both encompass assumptions about what the target uses are
likely to be and are optimized for such target uses.  If those target uses
are significantly different, it will always be possible to construct an
example shaped to fit the target for either Manifold IMS or MapServer where
one clobbers the other.   If that is true, the question then becomes which
target use pattern is more relevant to the specific use one is considering.
I believe that Manifold is better aimed at the target uses that predominate
in the market.

Manifold IMS is session-free and supports object pooling, both wonderful
performance gains in an .asp environment.  Microsoft ASP in itself is
obviously a higher performance environment than CGI (although, of course, it
has tradeoffs in other dimensions).  There are well-known technical reasons
why this is so that cannot be made irrelevant by labeling them "mythic" (as
you apparently acknowledge by your comment on "CGI overhead").

> is slow, MapServer is actually the fastest fully featured web map server
> I've seen until now... yes CGI overhead exists but it is minimal
> compared to the time to render maps based on hundreds of Megabytes or
> even Gigabytes of data (which MapServer can accomodate amazingly well
> thanks to its high performance internal architecture!).
>

It sounds like you are making the case that the well-known, less-optimal
process performance of CGI relative to ASP does not matter because it is
dominated by the slow response of the server when rendering images based on
very large data sets.   That's a fair point, but it is only true in cases
where very large data sets are in play.

It turns out that (as a determined web crawl will reveal) the maps that most
people publish over the web are tiny.  For the most part they are simple
vector maps running a few tens of megabytes.   I could be wrong, but I
believe that all other things being equal CGI overhead (if nothing else) in
such cases will be a negative factor in an otherwise well-architected map
server.  This will be especially true with many hits in a short period of
time.

On the other hand, "performance" is something of a canard since virtually no
map servers have any significant traffic.  For the most part the
distribution of map servers on the web is lots of niche sites that few
people visit.  This is a better argument in favor of CGI than the one you
made.


> Anyway, I'm looking forward to see a Manifold-based website that can
> beat Jean-Francois Doyon's Election application which served 80,000 maps
> from a single Linux box on elections day last fall while sites based on
> other webmapping servers all went down.  There is no cache trick here...
> we're talking about real maps drawn and served.  Combine that with a
> pool of servers with load balancing and the possibilities are unlimited
> (and the price hard to beat!).
>

Hmm... 80,000 / 24 = 3333.  3333 / 60 = 56 (rounded).  That's less than one
image per second.  Clearly, there must have been peak load issues if other
servers went down since a server rate of one per second does not strike me
as a very high standard in modern times.

"Cache tricks" are part of an intelligent design so they should not be
disparaged.  The only "real" map drawn and served is the image the user sees
in his browser... any effective method is fair game.  Plus, one can load
balance with any system through simple methods.

I disagree on price, by the way.  Few things are free: there is usually only
a shifting of cost.   To really "do" map web sites a user needs both a map
server and a GIS.  The cost of ESRI + MapServer is far beyond the $245 cost
of Manifold + Manifold IMS.  In both cases, the map server is "free."
That's the equation confronting most users (who are new to GIS,
statistically) who would operate a Internet map server site.

My own guess based on rather cursory explorations is that fewer than 1% of
the potential users of an Internet map serving site need more than one
server.  For them the above equation is dominant.  For those who need more
servers the benefit of a free map server increases since presumably they
don't need to replicate the GIS.  However, there is a substantial cost in
technical expertise imposed by MapServer.  That is not a free economic good.
Humans cost more than anything in the system, so when even a little bit of
thought is required the cost penalty is high.  If you price what it costs to
get a consultant in to configure MapServer for you, I'd bet you would
conclude that the actual cost per server for small numbers of servers is
higher than the volume discounted price of Manifold IMS for such servers.

It's true that if you have a server farm with 1000 servers you have a
serious incentive to get free software for each. But, as a significant
percentage of the market such installations do not exist.  It's also true
that some people have the intrinsic technical skill to configure MapServer
or the hobby interest to acquire such expertise or enough time on their
hands that the cost of acquiring the skill is zero at the margin. For such
people there may well be no "cost" to using MapServer but once again as a
significant percentage of the market such a cohort does not exist.

> 80,000 maps per day on a single server is acceptable performance to
> me... if Manifold IMS beats that then I might even consider digging out
> our old Manifold license and upgrading to see it by myself.
>

As noted above, 80,000 per day is not a big deal: the key issue is peak
load. I feel safe in saying that given a reasonably modern PC Manifold IMS
has the performance to handle over 95% of all sites people might have.  In
fact, it has so much capacity that we expect many people will go into the
"application server" market of offering to configure and host other people's
IMS sites, much as Blue Marble is attempting to do as a service business.
One server could easily host many dozens of sites.

For the remaining 5% in a few months we will offer the Advanced Manifold
IMS.  Advanced IMS is optimized for those few Enterprise users who are
working with a different class of application.

Cheers,

Dimitri