> - What cost in terms of cpu cycles required of the
> systems serving the disks. (vax7640).
First place to look for general MSCP performance implication is:
VAXcluster Systems Quorum February 1992
on pages 48-61 there are detailed graphs and analysis of
both Ethernet and FDDI MSCP serving and the effects they have
on the 6000-510 used for the measurements.
Then if you just want a rule of thumb, from the data I've looked at,
you can assume that it takes less than .01 VUPs per average-four-block
I/O per second on CI or DSSI, and less than .02 VUPs per I/O per second
on Ethernet or FDDI.
You can approximate your VAX 7640 at 35 VUPS for this purpose. That
takes into concideration that MSCP serving is interrupt intensive, and
therefore pretty much limited to the primary CPU.
> - What cost in terms of responce times for io's from
> served disks .vs. locally connected disks.
The cost in terms of responce time will be minimal, a few ms. The
total cost might be higer as you are using your expensize VAX 76x0 as a
cheap disk controller. You might want to plug a CIPCA in the
AlphaServer 2100 and have it talk to the HSC directly. It will leave
you VAX for what it was intended to do!