[ Home ] [ Table of Contents ] [ About Lee Goeller ] [ Search ]

It Ain't Necessarily So

I wrote this and the next Ain't in May, but Ian cut both of them down and packed them into a single page in the June-July 1992 issue of Telemanagement. Here they both are, each in its full glory.

Processing Cheerios

Many people have been successfully conditioned to demand the latest and most up-do-date kumquat squeezer on the market today, even when it does nothing their old kumquat squeezer doesn't do. There is just something about having the latest that drives people to buy. Even those who can read seem to prefer best sellers; Dickens is ancient history.

The automobile industry is, perhaps, the classic example. People rush out and buy a new car, whether they need one or not. My Mercedes Diesel is now 23 years old and going strong; I can laugh at them. But there are pressures to force even laggards like me to comply. For instance, to pass New Jersey's inspection five or six years ago, I had to replace a worn out LICENSE PLATE! And much as I am impressed with German engineering, my ignition key is about to fall apart. It is not only Detroit that uses subtle means to get you to buy a new car!

However, the computer biz has gone automobiles one or two better. We pioneers who led the way in 1978 were ultimately forced to shift from our 8-bit 8080-based machines to 16 bit monsters based on the 8086/8088 favored by IBM. Needless to say, we had to replace all our software, and converting a backlog of files was a real pain, particularly when we had used several 8-bit operating systems in addition to CP/M.

MS or PC-DOS, the de-facto standard operating system for 16 bit machines, sanctified by IBM, took over and provided a market to justify the development of a fantastic wealth of software, and also a boom in various peripherals. The backlog of files accumulated by anybody with a computer-based career continued to grow beyond all belief.

But there came a time when even the 16-bit market saturated. What to do? Fear not -- Intel to the rescue. The 32-bit 80386 chip appeared, and suddenly the marketeers could start all over again. The modest replacement market based on equipment actually wearing out was gone, and in its place was the opportunity to resell the entire existing installed base! Utopia!

But the library of irreplaceable user files, added to the value of the software that had permitted those files to be created, exceeded the value of computer hardware by orders of magnitude. To get the customer to buy new hardware, that hardware HAD to carry along the old software and files. And it did. Intel cleverly designed the 386 chip to use the same old PC/MS-DOS, and all the programs and files dependent thereon.

But the 386 is a 32 bit chip, and DOS is a 16 bit system. So what? Who says you have to use all 32 bits? The higher speed and greater memory access possible with a 386 more than justify its use, according to its more rabid fans, although each operation processes 16 bits of data and 16 zeros. Processing cheerios is the name of the (compatibility) game. And most PC users remain blissfully unaware of all that unused horsepower under the hood. Advertise the power, sure, just like the 400 horsepower cars we used to have before OPEC punctured the balloon. And just as there is no way to actually get 400 horse power to the wheels of a passenger car, there is no way to get 32 bit processing power to an MS/PC DOS program. But who cares? We just want a nice drive to the country and an easy way to write a letter to grandma.

PCs aren't the only devices that process cheerios. The Rolm/IBM/Siemens 9751 follows the same pattern. The original Rolm CBX was based on a 16 bit processor, and used a 16 bit parallel bus as a switching matrix. With 12-bit linear PCM coding, a 16 bit bus had four bits to spare, and worked fine. The bus was improved, renamed ROLMbus 295, and passed on to the 9751. However, because the 9751 used standard mu-law coding, only 8 bits were needed. The other 8 carry cheerios (except for an occasional internal control word that uses more than 8 bits).

Most packet switches use parallel buses (8, 16 or even 32 wires) to carry a lengthy burst of one or several from the source port past the all ports including the destination which, upon reading its own name in the header, copies the packet as it goes by. Packet switching differs from circuit switching in that one packet will use the bus for a number of consecutive time slots until the packet has been sent, while circuit switching assigns each connection one time slot for each direction of transmission and interleaves a number of different channels. A circuit switch can handle many channels simultaneously but at a relatively low speed, while a packet switch handles only one packet at a time, but disposes of it much more quickly because it uses the entire bandwidth of the bus until its job is done.

Rolm is in the interesting position of being able to add a 150 Mbps packet switch to the 9751 by simply replacing cheerios on ROLMbus 295 with something useful. Let 8 bits on ROLMbus 295 continue to handle circuit switching as at present, but turn the other 8 bits into a packet switch. So far, Rolm has ignored what seems to me a great opportunity.

Now that AT&T has set up its Definity G3r to use parallel bus switching for both individual line groups and the group selector, it will almost certainly add packet switching capability in the near future. Maybe this threat will spur Rolm into action, and we'll have fewer cheerios to deal with. But maybe not; in a highly competitive market, innovation is often too costly to risk. We are lucky that AT&T, during its monopoly years, developed a habit of innovation that hasn't yet been stamped out.

[ Top ] [ Next ] [ Table of Contents ]