<Will that be PIO or DMA? 

I left that open.

<Even though I'm not very happy with IDE it can do reasonably well 
<in a 1-drive-per-controller setup.

If cheap no problem.  Howabout several drive but each off a seperate
cable (non interactive).

<What I'm a bit worried about is the resulting speed from a PIO 
<design. What can we expect? VS3100 PIO speeds? 30 Kbyte/sec?

Unknown but likely better as you don't have to twiddle the 5380 bits 
to get at the data.  the interface to IDE is buffered enough that the
cpu would not have to wait for data.  That is not true for the raw 5380
scsi interface despite a large buffer on that may be on the SCSI disk.

<And how much of the VAX'es (precious) CPU cycles will it be
<using (wasting?) copying those bits from the drive?

Personally PIO is not the best but can be real cheap.

<DMA does rule out some older IDE drives which either don't support 
<DMA at all or are quite broken in that respect. Lots of cruddy 
<implementatin in the peecee world..

Personally with new 1gig drives at $99, older under 3-400mb drive are as 
far as I'd go back.  There is no point to that.  Especially when CompUSA
(no bargan house) sells 7gig drives for under $300 new.

<But as for SCSI..
<
<>From the discussion(s) here I see that SCSI would immediatey
<mean 'on-board CPU'. I'm puzzled by that..
<
<If you grab a 5380 derivative (preferrably better ;-) it should 
<only need some logic for DMA'ing, but the chip should handle
<most of the low-level SCSI stuff itself.

The vax (for 5380) would have to twiddle the bits to do DMA setup and 
post transfer clean up and a fair amount of hardware handling inbetween. 
The 5380 is really crude and does require a lot of CPU attention to keep 
the SCSI bus real.  I've worked with the ncr5380 in z80/z180 systems and 
it was no fun, besides it's old enough to be out of production.

<I use PC SCSI controllers (eg. Adaptec 1542 ISA with bus-master
<DMA) as a reference here.. These cards (almost) never use an extra
<on-board CPU.

It may be hidden as a house marked custom micro like 8051, z8 or any one 
of a dozen others.  Also I dont use PC cards as valid examples.  The 
older cards like 1542 may use the cheaply available cpu cycles to hadhold 
the scsi bus controller.  

Also DMA on qbus is limited to the 22bit(4mb) address but I have no idea 
not that maps to a MV-II with 16mb!  this would affect both scsi or ide.

Ideally in the case of SCSI it's nicer to hand over a higher level 
command packet and leave it to the local controller to twiddle bits 
and do the DMA.

Allison