Christopher B. Browne's Home Page
cbbrowne@acm.org

18. Network Computing

This was the "hot new thing" of 1997...

The general architecture is that instead of the common arrangement today where users have a general-purpose WinTel PC on their desk, storing some data on LAN servers in the back room as managed by users (that probably don't have a clue of how to manage the software and configuration on their computers), we have a smaller, diskless workstation on the user's desk which hosts some applications, and leaves others to actually run on the server in the back room. All data resides on the back-room servers.

It isn't as new as the publicity would suggest; the basic architecture was implemented and commonly used probably starting ten years ago using "X Terminals," where the "central server" was a network of fairly large Unix boxes, and the workstation was the "X-Term," a terminal primarily just smart enough to run an X server to display applications using the X11 Window System. The recent change is to add protocols specialized towards hosting the display of Microsoft Windows applications.

The first "Network Computer" implementors were HDS and NCD, companies that have been building and selling X terminals for a number of years.

The major difference between Network Computers and traditional X-Terminals is that NCs add a bit more software to the desktop box so that:

These changes reduce the amount of data that needs to go through the network connection, the amount of computing power required on the server, and associated memory consumption on the server. The other significant change is that the overall system includes the capability to run MS-Windows applications on a central server and export the displays to the desktop NCs.

There are plans to provide some additional protocols for which servers would reside on the desktop box, such as a special protocol called "ICA" to allow display of typical Windoze apps that use Microsoft's "GDI" API. The application would, again, run on the central server, and be displayed on the desktop box. (Apparently Microsoft disapproves of the notion of doing this using the X protocols. Big surprise... I wonder if anyone will be able to write an "ICA" (Intelligent Console Architecture) client that can talk to X servers, thus circumventing this?) Recent work involves using Java-based applications to manage the export process.

This approach allows desktop machines to become much smaller and cheaper, as they will typically not need all of the wide array of peripherals that appear on the typical desktop PC. Typically, mechanical components including disk drives of the various types (CD, Floppy, Hard Drive) go away. With these items being removed, the desktop unit shrinks in size, price, and complexity of installation. For instance, IBM's early NCs were about the size of a paper notebook at 20x25x3cm, and weighed only 1Kg (about 2.2 pounds). And the insides of the unit were not tightly packed... It doesn't need to be even that large, although there is little value to making the unit much smaller than that.

American Airlines rolled out some of these Network Computers for airline maintenance centres.

The removal of mechanical components also improves reliability, and ventilation using convection may be quite sufficient, if one's desktop machine doesn't involve hardware that nearly requires water cooling.

A typical Network Computer doesn't need to be larger than a video game console; it would be a neat thought for Nintendo or Sony to turn one of their video game consoles into NCs. They ought to be able to sell reasonably powerful NCs for under $400.

Particularly valuable about the approach is the notion that you no longer need to configure PCs individually. Your configuration resides on the network, and can follow you around to whatever NC you use. If something breaks, it may be an entirely reasonable idea to "fix the PC" by throwing it away and replacing it with a new $400 box.

18.1. Pros and Cons of Network Computers Versus "Wintel PCs"

Google
Contact me at cbbrowne@acm.org