My memory is that the “multiplexor channel” could handle a huge number of
simultaneous transfers, doing the equivalent of DMA from multiple
high-bandwidth devices. High-end machines could do significant fractions
of a gigabyte per second. I vaguely recollect that memory was
multiported, but that may have been one of the many custom mods our
machine had.
Gordon Bell, one of the founders of Digital Equipment Corporation (DEC)
and the architect of the PDP-11, left DEC and spent some time in the
supercomputer industry. Gordon loved to crunch numbers. He found, for
example, that the National Science Foundation grants for access to
supercomputers were typically around US$5000. With this budget, a typical
researcher could afford one minute a day of supercomputer access, and most
programs ran in about a minute. So the researcher got one result per day.
In 1990, a 386/387 system would take 24 hours to run the same program,
and this meant one result per day. So researchers were spending $3000 to
get a 386/387 that could run their problem, and with the extra $2000
getting a second, somewhat less powerful, machine for word processing,
spreadsheets, and (remember, this was 1990!) sometimes email. So they
were going after easier-to-get grants of $5000 that gave them not only
their one result per day, but other benefits their supercomputer grant
couldn’t give them.
Now compare that to a Xeon or Core series, and we can now buy, at our
local retailer, more power than the supercomputers of 1990.
The bottom line is that the supercomputer market is limited to a few, very
few, customers: weather forecasters, nuclear weapons simulators,
cryptology, and…well, there just aren’t a lot of people with big
problems and big budgets to match.
joe
In around 1991, my friend was using the latest Soviet /360 (or /370?)
clone to run his Fortran scientific computations.
When he switched to 386 with 387, he said that the perf was similar, and
with Weitek FPU the modern (to the date) PC was beating the old (old for
US, not so old for USSR/Russia) mainframe.
> throughput. A 360 equipped with a multiplexor channel
Was this something like a chain DMA-based storage controller?
> performance from the bulk memory. The result was the whole machine went
> down for a week
–
Maxim S. Shatskih
Windows DDK MVP
xxxxx@storagecraft.com
http://www.storagecraft.com
NTDEV is sponsored by OSR
For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars
To unsubscribe, visit the List Server section of OSR Online at
http://www.osronline.com/page.cfm?name=ListServer