On Thu, Sep 04, 2014 at 05:13:08PM -0700, Al Kossow wrote:
My best guess -
given what the seller's indicated - would be that the
QED's were used in an existing ticker plant or market data system,
handling leased line serial comms on one side, using the RAM disk to
buffer data or speed up queries, and just horked the data around without
acting on it. If it's just packets or blocks of data being fed to other
systems, the lack of floating point wouldn't matter so much.
John Wilson knows the details, since they were also one of his customers
according to Dave Carroll.
I forget in what ways I'm sworn to secrecy so I'll keep it vague, but if
it's the customer it sounds like, they were running all proprietary code
on 11M+ and using DECnet for the transport. Trades were load-balanced
across hundreds of PDP-11s (by splitting up the alphabet I think?). It
was frickin' AWESOME, and a religious experience to visit. All eventually
moved to Linux (which was an easy step once the emulation was on Linux,
and it ran as a hybrid for a while with some pieces inside the PDP-11
and some outside using the NP: named pipe interface which I'll document
some day).
Apparently several companies, including Mentec, were
supplying boards to
keep these systems getting faster, and the sources would change depending
on which vendor had a faster board for the application.
Those guys were wonderfully pragmatic. They used whatever was fastest but
always kept re-evaluating whatever else was out there (so yes they used
QED and Mentec and I think also Strobe hardware, before E11). They didn't
believe marketing claims from *anyone* and tested everything heavily on
their own workload before deploying it. I got some of the most fantastically
specific bug reports from them ... the kind where you fix the bug in five
minutes because they tell you precisely how the behavior is wrong.
John Wilson
D Bit