| Fig 3: Card with 6 transceivers
|
Fig 4: Multi- card lattice in 1 plane
|
Fig 5: Multi- card lattice in 3 planes
|
Each yellow rectangle in the above illustrations represents a circuit board (like a motherboard) with a CPU and RAM along with 6 transceivers. Using bi-directional transceivers mounted on a circuit board facing in 6 directions (as in Fig 3), it is possible to create part of a computer lattice. Such a lattice could grow in all directions, allowing very large numbers of processors to operate as one system. The direction of communication in 3 planes allows information to flow easily through the system, and makes it possible to re-route information if a link becomes damaged. The red and green colours in Fig 3 show where the red and green transmitters are placed on the card. Fig 4 shows a series of cards in a simple lattice with transmitters used on only 1 of the 3 planes. By following Fig 4 and Fig 5, it is easy to see how the lattice network could expand in every direction. Although the cards are shown spaced apart (to make the links visible), it is likely they will be end to end in practice. This would make 4 of the 6 transceivers almost butt together for very short transmission distances. Only those placed above and below each card are expected to need fibre optic cable joints. Using this method, computers with thousands of parallel processors could be quickly and easily made. These boards would be designed for rack mounting, so many could work together. A suitable control system for the optical transceivers would be essential. Possibly a combination of DMA with UART along with optical sensing devices would handle this need. If any such exists, the author would like to hear of them.
Wouldn't a computer like this be a nightmare to wire up?
NO! In Fig 6, below, the main wiring scheme is almost too simple. The boards are slid into position between two metal rails, carrying 0 V and +5V respectively. Contacts on the board edges pass power from the rails to the circuits on the card. All other signals are passed through the bi-directional transceivers. There are no other essential wires needed. Any variations in on board voltages may be produced from on board regulators. Also, by designing the rails as extruded heat sinks, most of the heat generated on each card could be dissipated (provided it could be conducted to the rails by the cards). NOTE: It would be likely that passing signals through holes in these rails would be an effective way to send information on the plane parallel with each card. For that reason, it is suggested that holes be drilled at suitable intervals along each rail.
| Fig 6: Wiring up the cards |
|
What about peripherals?
Suppose I want a printout? Current PC systems are very peripheral oriented. So, by creating an optical fibre connection to a card on a PC, peripherals may be shared as they are on a conventional network.
What if something went wrong, and you had to reset?
Each board would normally reset on power up. To reset a bank of cards, a switch may be fitted to each 5V rail to momentarily disconnect and power up the bank. Other banks would not be reset, so normal operation could continue while an affected bank is brought back into service.
How would you replace faulty cards without taking the system out of service?
Just disconnect any fibreoptic leads from the card and slide it out of the top of the bank. The others will not be affected electrically, although some information may have to be re-routed. Slide in a replacement, and reconnect the fibreoptic leads. If necessary, one of the 5V rails could be switched off, while cards in that rack are changed.
Where would the operating system be stored?
This system would use an EPROM based BIOS designed to download any desired operating system across the network on reboot. The OS could be stored in a shared directory on any PC with an optical network card. The BIOS itself would support a RAMDISK for needed files. Existing large DIMM memory devices make this much more practical than it was a few years ago.
How would it provide for user input/output?
On boot, the BIOS queries all network ports and registers those with PC features (keyboard/screen etc) so that all input received from that port can receive response to the same port. In this way, a user could type text to the network and see it appear in a screen window as the network unit displays it.
How would this system differ from current systems?
Lots of ways:
1) Bi-directional transceivers pass data fast without electrical connections.
2) No complex electrical connections means no earth loop problems, so few power disruptions.
3) It also provides very high immunity to lightning strikes and other dangers.
4) Can keep working even if many of its processors are damaged (Borg defence).
5) Large numbers of processors dedicated to processing information instead of supporting many peripherals means massive computing power, easy multitasking and many user connections.
6) The computer can be expanded almost indefinitely (as long as there is power to run it).
7) Simple to build and maintain (put in another processor and keep going).
8) More powerful processor cards easy to include as they are developed.
9) No physical size limit. Size is limited only by BIOS and operating system (infinite PC).
10) Ideal for massive internet servers or parallel processing research or smaller scale work with only a few processors.
11) BIOS designed simply to reboot from the optical transceiver network and load any suitable operating system.
12) Entire system can boot from a single hard drive.
COPYRIGHT STATEMENT: This concept was originated by John Spencer Durham in June 2000, and posted on the internet in July 24 2000. Anyone with the means and desire to develop this principle into a commercial system is free to do so providing the originator is acknowleged by name along with the product, its packaging and manuals.PREVIOUS..............................NEXT
