GIS and Networking

"A little learning is a dang'rous thing." - Alexander Pope

 

Few modern phenomena have taken hold as rapidly as the recent fashion for networking technologists to make money by taking advantage of the greed and stupidity of stock market investors. There's no better way to start a stock market craze than by leveraging an exotic new technology or business method that the average investor does not understand. Add some pseudo-scientific gloss and you can take them to the cleaners every day. Even after a dot-com bust there are plenty of investors lined up to pursue the next big new thing.

 

The opening years of the new Millennium are an especially fine time to fleece investors with new Internet schemes or network-related glitz. At the time of this writing much of the world is riding a wave of immense prosperity compared to previous decades. Even with stock market declines, lots of people still have lots of money to invest. At the same time, math and science education in the US and many developed countries is at an all-time low. Never before in history have so many poorly-educated people had so much money they are willing to hand over to confidence men who can dazzle them with big words and mighty promises.

 

"GIS over Internet" is the GIS industry's own contribution to fleecing the ignorant. By this phrase we don't mean publication of images but rather the idea that GIS in local machines will be replaced by GIS done mainly on applications servers working with dumb, local clients.

 

As an Internet company that sells Internet Map Server software we would be the last to suggest that Internet is not in fact a revolutionary new way of communications. To compare Internet to the invention of movable type and its popularization by Gutenberg may actually be an understatement. Internet and the technologies it spawns will in our view result in a greater transformation of human society than the appearance of fast and easy printing technology.

 

However, just as the simple act of printing does not guarantee infinite wealth, just so simply using Internet does not guarantee sustainable wealth. To continue to make money in the long run one must use Internet sensibly. So as not to accidentally invest in really dumb ideas that are being promoted to take advantage of the Internet craze we have to understand some of the technology involved. Investing based on technical reality will help us keep our eyes open during market run-ups drive by speculative idiocy and will also serve us well in avoiding losses when investments based on fantasy collapse. Even if investing in stocks is not the game, understanding the fundamental conceptual errors behind GIS over Internet schemes will help organizations avoid the expensive mistake of investing in such schemes for their own operations.

 

Processor Speed and Internet Speed

 

Local area networks have continued to increase in speed to keep pace with higher speed local disk accesses. Small businesses and homes have shifted to hundred-megabit speeds and are beginning the move to gigabit networks while large businesses have already shifted to gigabit networks. People are using the speed of these local networks to make it much easier to reach out and get data regardless of where it may be stored. Modern local networks are therefore great at creating a transparent group computing experience when it comes to disk accesses.

 

However, disk accesses are only part of the picture of what makes for a modern computing experience. As fast as modern disks are, they are a thousand times slower than the fast parts of the computer. Hard disks typically work a few thousand times slower than dynamic RAM. RAM in turn works many times slower than the processor driving the system. Comparing the speed with which a processor talks to local RAM and local video memory to the speed with which a hard disk operates is literally like the speed difference between a fast bullet and a slow human walk. See the Using RAM and other Machine Resources topic for what this means to an application user.

 

This speed difference is an important factor because the human perception of computer response (whether we judge an application to be fast and responsive or slow and unusable) is largely driven by what happens at processor, RAM and video subsystem speeds. The practical limit on how cool or intelligent or responsive a modern user interface may be is largely determined by the speed of the high speed parts of the computer. This is something every 14 year old knows who has stepped up from playing a game on a slow video card on a slow processor to rocketing along with a fast processor and a big-time graphics engine with lots of megs of video RAM. The speed of the hard disk or network has zero influence on the quality of the immediate user experience, because both are far too slow to participate in the user experience even in the slow case.

 

This speed differential calls up another factor that many GIS over Internet and Network PC investors fail to understand: As slow as hard disks and local networks may be compared to the bandwidth between processors and local RAM, Internet is many, many times slower. Consider that all the exciting talk in recent years about "broadband" Internet connections boils down to pathetically slow speeds, on the order of a megabit per second or so, a thousand times slower than a fast local network and usually millions of times slower than local system internal function.

 

If the difference between processor - RAM speed and hard disk speed is more than the difference between a speeding bullet and a human walk, then the difference between hard disk speed and Internet speed is literally like the difference between a human walk and the slowest crawl of a snail. Even a very fast connection to Internet is literally a thousand times slower than a local hard disk, and it is approximately a million times slower than local processor - RAM accesses.

 

Forgetting this key technical difference leads to business mistakes that are as dumb in their own way as the decision of the Tennessee state legislature one year to vote to make "Pi" a round 3.00... it just doesn't work. The plain facts of the matter are that Internet speeds are a million times slower than speeds achieved between a processor and its local memory. More importantly, although the speed at which processors and local memory communicate are getting faster every year Internet speeds are growing in speed at a much slower rate of improvement. This means that the arguments presented in this essay become stronger every year, not weaker. Quite soon, as processor - RAM speeds continue to rapidly increase while Internet speeds hardly increase at all, the difference between "in the box" speed and Internet speed will be like the difference between a fast bullet and a fast-growing plant.

 

The Uses of Speed

 

The main reason one needs speed is to support a high quality, modern user interface. A fast, elegant user interface tosses billions of bytes back and forth between processor and RAM to make the user experience smart and easy. Manifold System, for example, runs thousands of functions as a mouse drifts across a map. Every one of those functions depends on visual positioning and other factors to make the user interface more intelligent. Because human time is much more expensive than processor cycles it makes sense to throw processor cycles by the billions at even small tasks just to save a little bit of time for humans. The computer should have to work very hard so humans need not work so hard or study as much.

 

Even the fastest Internet connection is thousands of times too slow to support such sophisticated user interfaces. Remember that a processor - RAM connection is like a bullet to a walk when compared to a hard disk, and the hard disk is like a walk to a slow snail compared to Internet. Folks who make the technical mistake of thinking that Internet can support the same user interfaces as a processor - RAM connection are making as dumb a suggestion as it would be to suggest that a snail could play on a World Cup soccer team. Actually, it's worse than that. They are suggesting a snail can keep pace with a fast rifle bullet.

 

This is the main reason that GIS over Internet cannot handle the modern interfaces expected of modern applications. Internet is simply too slow. This is also the reason people do not do Adobe PhotoShop or other graphics editing across Internet, nor do they do even low-bandwidth tasks like word processing over Internet. Internet is just too slow. This is something everyone knows who has any actual experience of "Internet applications" such as web-hosted email. Such applications feature frightfully slow and stupid user interfaces that are hampered by the very slow speed of Internet. [They are also hampered by the "tragedy of the commons" effect whereby Internet connectivity is always automatically overloaded, but that's a different essay.]

 

GIS across Internet proponents often respond to this point by making a really stupid comment. They point out that Internet speeds are increasing every day. That's dumb because the comparison is not between Internet speeds of today and the speeds that Internet might achieve tomorrow. The comparison is between Internet speeds on any given day and the speeds achieved between processors and RAM "within the box" on that same day. Internet is getting faster, but processors and "in the box" speeds are getting faster still. The comparison is not to a fixed target, it is to a moving target driven by relentless speed increases in the link between processors, RAM and video memory.

 

Sure, modern DSL or cable modem Internet connections are fast enough to support user interfaces that are better than those on time-shared, Teletype printers or plain vanilla ASCII terminals as were common in the early 80's. So what? People are no longer satisfied with such interfaces. The requirement is not to grow Internet speed so that one day it will equal the obsolete user interfaces of the past. The requirement is to equal the elite user interfaces of modern times. It is a race in which Internet speeds are falling farther behind.

 

This is the key technical disconnect that wishful thinking and glib marketing often hide from the unwary. As technologists, we fully expect Internet and Java applets and such to get a hundred times faster in the next five years. Big deal. That's still a million times slower than the bandwidth achievable today by using local processors and local memory, and the Internet speeds of five years from now will still be thousands of times too slow to do a really first rate user interface by today's standards, let alone by tomorrow's standards. If one expects the standard for "power" user interfaces to improve in the future, then future Internet speeds will be even less suited to serve the user interface standards of tomorrow.

 

One main reason that Internet speeds are not increasing as fast as processor and memory speeds are increasing is that the speed of processors and RAM is driven mostly by local advances in silicon lithography and fabrication that are unconstrained by large scale infrastructure. Any improvement in any silicon fab quite rapidly circles the globe. In contrast, the speed of Internet can increase only as large scale infrastructure can improve while simultaneously local infrastructure and technology must also improve. A slight reduction in line width in the fab automatically yields much faster chips. In contrast, even slight increases in Internet speed require massive changes in physical infrastructure due to constraints imposed by basic physics. Rewiring the communications infrastructure of the entire country to go from copper to fiber is massive enough, but even that would be a million-fold slower than current advances in silicon fabrication can provide.

 

A final restriction on the speed of Internet for interactive computing processes is the speed of light. Already, computers are so fast that speed-of-light issues are beginning to arise even over the short distances within a computer. The physical distances between a client and a distant Internet server impose a crushing limitation on speed as a result of speed of light travel time between the two, over distances that are millions of times farther than intrachip distances. Although this may seem to be science fiction to the technically illiterate, something that might be erased with a suitably glib disclaimer in marketing documents, the speed of light limitation is a genuine factor in high technology that cannot be wished away.

 

In all fairness, Internet is fine for "batch" operations that are not interactive. It's also fine for presentations and light-duty user interfaces. However, the issue is not the complexity of "batch" spatial operations that can just as easily be done on a server as on a local workstation. The issue is the quality of a sophisticated, interactive user interface, which is far more bandwidth intensive. One can easily write a fearsomely complex query in a few text lines of spatial SQL. The point of modern GIS is avoiding the use of dumb text interfaces for such things whenever possible and instead substituting a visual, seamless, point-and-click approach. Again, the point is to use infinite machine cycles to save us humans from having to do drudge work.

 

We should also note that giving up one's GIS solution to an off-site server is a sucker trap that defies human nature. It is simply asking too much of human nature to believe that any server operator will not over-sell and under-power the server. As anyone who has ever used Internet knows, they don't call it the "World Wide Wait" for nothing. In modern times a multi-gigahertz processor is so cheap that one now buys such supercomputer-class machines for one's children to use on schoolwork. Who would want to timeshare any machine when we can have our own personal supercomputer for lightning response whenever we want it? The buzzword for this in organizations is "scalability" - if each new user brings a supercomputer class machine to the party, we can keep adding users as long as we want without performance problems.

 

We think Internet is great for selling product, for delivering data sets, for synchronizing work groups, for publishing summaries of work done and delivering updates and all sorts of work that does not require an intensive, interactive user interface. Products like Manifold's Internet Map Server capabilities are great for such things. However, if one wants a smooth, intelligent, sophisticated, fast, and assisted user experience, it's clear that Internet-based applications will lag far behind modern technology applied "in the box."

 

Willing Slaves

 

So, why does one hear so much about GIS over Internet? We think it boils down to those two eternal determinants of all technology: politics and money.

 

The politics are partly about Bill-envy and a desire to outflank Microsoft. The ill fated push into Network PCs (billions of dollars down the drain on a really dumb idea) by an alliance of Microsoft competitors was motivated mostly by a desire to wedge a crowbar between Microsoft and users. The idea of a Network PC is that all intelligence and power should reside in a central server, which serves web pages to low-power machines that do nothing but operate network clients.

 

The Network PC is simply the multi-user idea of a central processor and dumb terminals brought back from the dead. It is recasting the UNIX dream of a universal, quasi-dumb X terminal attached to a server into a modern lexicon of "Network PC" and "Application Server." The product-marketing basis for pushing Network PCs was that it is easier to pry users away from Microsoft if the only application that needs to be ported is a browser. The Network PC failed, of course, because neither PCs nor their users are stupid. The benefits of becoming a willing slave to a central organization are far outweighed by the benefits of distributed, independent use of one's own intelligent machines.

 

Another set of politics pushing for "Internet applications" revolves around big IT organizations discovering a fine new tool to centralize everything once more. It may or may not be convenient to have someone else do system administration but it is seriously convenient for organizational empire building when all computing activity is centralized within the organization that runs the servers.

 

The money factor in GIS over Internet has to do with the magic of selling the same thing piecemeal instead of as a complete data set. About 30 years ago rare book dealers discovered that a hand-written book on vellum that could be sold for a few thousand dollars as a whole book could be sold for tens of thousands of dollars to a much broader audience if the individual vellum sheets were razored out of the book, mounted in a pretty mount and sold off one by one to collectors. Within a few years, an entire generation of manuscript books disappeared onto the walls of middle-class collectors as framed, single page ornaments.

 

Serving single images of maps over the web earns more money for database owners (either in a pay per view or as a lure for advertising) than selling the maps as complete digital databases. It's also a great way for the owners of digital map databases to retain ownership of the real GIS content. For public agencies who have come to enjoy selling public data for high prices it's a great way of being able to say they are meeting their mission of making public data available while in fact denying the public access to the real data. This allows the agency to charge high prices for data on CDs. It also allows bureaucratic expansion of programs and personnel needed to "develop" web pages instead of the relatively simple task of publishing data sets.

 

Another important factor in the money side of the equation is that there's a lot of financial pressure on traditional GIS companies to transform themselves into something other than a "old economy" software vendor. It can't be easy for someone who has spent decades building a GIS company to see a group of twenty-something web entrepreneurs achieve a valuation of a billion plus dollars in less than a year, merely because they thought up a better Ponzi scheme involving a dot com, or an LBS play or wireless play or whatever. Even with stock market declines, companies that can put a network spin onto their business are valued more in today's stock market than are ordinary software companies. Selling oneself as a "Maps over Internet" player is a great way to increase valuation.

 

Selling high and mighty web service ideas to technically illiterate investors is also easier work than competing with aggressive fanatics in software markets. If you were running a traditional GIS company, what would you prefer: fleecing Wall Street with web / wireless talk or risking financial ruin by getting into a price war with the likes of manifold.net? No wonder some GIS companies have promoted "GIS over Internet" like their lives depended on it.

 

We would like to point out that some institutional users with the best of intentions trick themselves into thinking GIS over Internet is a cost effective idea based on the notion that a browser costs nothing while a real GIS package can cost a lot. If the cost comparison is against some ancient ESRI software item that costs thousands of dollars a seat we can understand how such a cost comparison might be persuasive. However, if the cost comparison is against something like Manifold, which is inexpensive to begin with and heavily discounted in volume, then the idea of constructing an expensive web-interfaced GIS is a false economy compared to deploying Manifold on client desktops.

 

The web-based GIS will have a poorer interface than Manifold, so the human cost of actual use will be higher (immediately eliminating any savings of using free browsers compared to Manifold license fees). A more directly visible cost will be the cost of developing, deploying and maintaining a semi-custom web-based GIS system. This cost will greatly exceed the cost of Manifold licenses. Worse still, the Manifold licenses are a one-time cost while the maintenance and continued development (after all, we do want to keep up with progress in software) of the web-based GIS will be a massive recurring cost in every annual budget.

 

The Manifold Way

 

At Manifold we believe that a cool user interface written by experts that takes maximum advantage of "in the box" power will always far outperform a GIS over Internet solution written by equally smart experts. For serious GIS work the best user interfaces and the fastest, largest and most sophisticated spatial work will happen within your local workstation.

 

Web interfaces are fine for casual use to serve up "pay per views" and other visual summaries of GIS work done in the box. Internet is fine for such things and is a good way to provide reports and pictures of one's work to a worldwide audience. It's also a great way to publish a project that allows simple user interfaces for browsing and elementary queries. That's why we have an Internet map server within Manifold: for such uses it's a great idea.

 

For real work with real data, even simple work with simple data, corporate users will always have the best experience using a GIS like Manifold locally even if the data is remotely archived. Architectures like Manifold System Enterprise Edition allow users to gain the archival and control benefits of centralized data warehousing while retaining the performance and local response benefits of truly distributed computing, where the application runs locally "in the box."

 

It is also important that GIS consumers use their experience to argue against schemes by public agencies that bottle up public GIS data behind a "webstacle" that merely serves images. Don't stand by while someone says spoon-feeding a few images is adequate public access to public data. Point out that if the GIS data itself is not made available, giving a summary report in the form of a web-served, dead image is not an acceptable substitute.

 

There is no corporate IT argument for GIS over Internet that does not apply with equal force to other corporate asset bases and activities such as graphics arts, images, documents, spread sheets and other corporate data. Whenever you encounter a "GIS over Internet" proponent, take a moment to ask them why GIS over Internet is such a good idea when PhotoShop over Internet or Microsoft Word over Internet or Excel over Internet is such a lousy idea.

 

Above all, remember the primary issue is not whether GIS over Internet is a good or bad thing. The key issue is whether it is worse or better than expertly implemented alternatives. It is also important to think clearly and not be fooled by whatever is the latest innovation in fleecing investors. See, for example, the What about Ajax? topic.

 

See Also

 

What about Ajax?

Using RAM and other Machine Resources