Now, I have proven to the satisfaction of everybody
that, ASSUMING
synchronous buses that are identical in every way except speed
(one being twice as fast as the two half-speed busses) that in two
real-life situations, one faster bus is BETTER than two separate busses.
In the real world there are to many conditions on things to make general
assumtions that always work. For simplicity and cost as a generalization
one fast bus will meet those needs and give good performance.
Now a counter to that. For error/fault tolerence, and if the fastest bus
is not fast enough then two (or more) will always win but will not be
cheaper.
I don't need number to prove that, the logic is deeper. To tie that to
the "old" machines they ran their generations PCI and then some. I'll bet
in a year or two you make the exact same arguement about the obsolete PCI
bus. To make a machine I'd consider to be in the same class as the 6000
was when new right now I'd expect to see all of the fastest busses in use
and multiples of them for parallelism.
Current generation PCs most of us have access to are not this case and do
perform well below the processor chips capabilities despite AGI, PCI,
SCSI++ (LVUW), copper giga and fibre. Because even the simplest process
like routing a packet requires the cpu to look at the address and see if
it belongs to the port A or port B list and all the DMA in the world
doenst get around that.
A good example is the DEC ugly(by some) RQDX MSCP disk controller.
While Qbus is maxed out around 4m words a second this controller can do
one thing to keep the lowly vax off it. It can DO DMA from LINKED lists
so that the controller is doing queued IOs. Now thats low end 1987
MicrovaxII technology but if we scaled that and put it on the PC it could
be keeping the cache filled and other tasks while all the cpu has to do is
set up task lists for it.
Another is this RAID thing, a good idea if you have SCSI. The VAX
was doing it over 14 years ago as part of the OS (disk shadowing).
OK so your 486 can do interger math faster, few said it was false. It's
just not enough of a measure. Less so to those guy where processing is
putting the right data (or part) in the correct bin and charging off
10,223.245 (yes three digits!) each for them. It may be meaningless
to the weather service that runs models that have data matricies that can
fill gigabytes of ram (which has to be filled first!) from their terabyte
disk farms.
All those MIPS, MFLOP and Dhrystones are wasted if the system structure
is hung up waiting for the floppy.
If your are preaching newer is faster, the choir is over there, they know
the words. If your saying PCs are better, well, put some bounds around
that as I really feel it's one of those ALWAYS/NEVER generalizations
that often bite.
Allison