I originally asked about why I was getting interruptions to my real-time
data acquisition system on an Alphastation 5-500/333. Thanks to Dr. Tom
Blinn for his always informative responses.
It looks like workstations may not be the ideal platform to be running
real time operations, especially data acquisition. The architecture in
general is geared toward graphics performance, so the graphics hardware
and software tends to have the highest priority running in the system.
On the hardware level, the graphics card has the highest bus priority for
interrupts and data transfers. That's why something as simple as a
mouse movement can interrupt our data transfer. On the software level,
what was a screaming fast CPU is now saddled with the chores of X services
and CDE operations, which involve many layers of very inefficient code.
That's why something as simple as a mouse movement can take away enough
CPU resources to totally foul up a real time transfer, even when the
aggregate transfer rate is not very high. I seem to remember running
into similar problems 10 years ago when we were trying to do the same
things on a Sun SparcStation. After 10 years of computer evolution, the
same problems still exist. How I long for the days of PDP-11 and small,
efficient kernels.
But since we don't have the time to build a PDP-11 based acquisition
system, our current solution is to pull the graphics card out of the
workstation and make it use the serial port for console. We will
connect to it using an X terminal, as we still need the graphics
capabilities, but all the graphics hardware activities will be
completely separated from the workstation. This is all theory right
now, as I have yet to do it.
eyc
Received on Fri Oct 23 1998 - 08:29:16 NZDT