HomeIndustry Commentary

The death of the Network Computer (first appeared in InfoWorld)

Like Tweet Pin it Share Share Email

According to Bernard Grun’s awesome The Timetables of History, King Herod died in the year 4 B.C. Christ was probably born the same year — if not he was born earlier, of course, since it was Herod’s call for a census that sent Joseph and Mary on their trip to Bethlehem.

All of you who still get riled up about when the millennium really starts should refocus your energy on fixing how we number years. Since the millennium began 2001 years after the year “-4” — 1997 AD if I’m doing my sums right — today’s column is really in the Oct. 26, 2002 edition of InfoWorld. The future is now.

Late in the real year 2000 I ran my first column trashing the Network Computer. Several years of marketing nonsense have muddied definitions almost beyond repair, so let’s try to restore some clarity to the situation: The NC, as defined by Larry Ellison, who coined the term, is a networked device that can execute Java code, connected to servers that download Java applications for local execution.

If local storage were expensive and bandwidth cheap, the NC would have made lots of sense. As it is, the whole attraction of the NC depended on two assumptions.

The first is that Microsoft will continue to avoid its DLL obligations. If you haven’t figured this out yet, Microsoft either created DLL hell deliberately or is so awesomely incompetent that our language lacks the words to describe its ineptitude. If Microsoft were to require registration of all DLLs and publication of their exact specifications, new versions of DLLs would not change functionality and DLL hell would be gone forever.

Of course, so would Microsoft’s ability to break competitor’s applications through the publication of new versions of DLLs, which is why I’ve concluded that this is the result of malfeasance rather than incompetence.

That’s one assumption. The second is the mirror of the first — it assumes someone would register all Java applications and applets, requiring that they all have fixed, published specifications. Otherwise we’d simply trade DLL hell for applet hell, and the sole advantage claimed for NCs — reduced cost of ownership — would vanish from the equation. Nobody has taken this essential step, and it’s pretty late in the game for the NC’s proponents to figure it out.

The IS Survival Guide didn’t have the clout to kill the NC — not in the real year 2000, not now.

Oracle has that clout. Since Oracle invented the concept of the NC, when it abandons the idea it’s safe to declare the NC completely dead. And the new version of Oracle’s ERP suite makes it clear the company has lost interest.

The new version, according to Oracle, is exciting because it migrates functionality back to big, centrally managed servers. It abandons client/server computing for a browser-based interface with all logic executing on the Web server.

If you like the idea of substituting WAN reliability and performance for replication, and you think Oracle has built a rich enough GUI into a browser, go ahead and buy it. Don’t, however, fall for the mistaken idea that if something is browser-based then it fits the NC model.

The point of the NC is for code — downloaded or cached — to execute on the desktop. The point of a browser is to provide an intelligent GUI presentation for code executing on a server. Yes, the browser can provide a home for Java applets, but Oracle’s centralized architecture isn’t suited for doing much on the desktop. That takes bandwidth, which means local, not centralized, servers.

Defenders of Oracle and the NC will rightly point out that the NC can host a browser, so Oracle’s ERP suite is compatible with an NC. That’s true, but no more relevant than the ability of the PC architecture to run Java Virtual Machines (JVMs) for hosting applications intended for NCs.

What’s relevant is that Oracle’s ERP team ignored the NC architecture in building this release. It isn’t built around Java applications downloading to desktop JVMs for execution.

And if Oracle’s ERP team ignores the NC architecture, who exactly is supposed to pay attention to it?