> On Apr 20, 2022, at 1:37 PM, Chris Zach via cctalk <cctalk@classiccmp.org>
> wrote:
>
>> Maybe a different guy, since Pro development was in the late 1970s to early
>> 1980s.
>> paul
>
> Probably. Still would be interesting to see what the board actually is/does,
> I wondered why they never did DMA transfers for the Pro, maybe being a single
> user system it was pretty much pointless or something. But the drives on the
> Pro are pure slugs compared to my 11/83+ESDI disks, they seem to be more the
> speed of an RL02.
The Pro drives are MFM drives, which are not necessarily all that bad. But the
controller is really inefficient, and you can see this from the fact that Pro
disks are formatted with 4:1 interleaving, so multi-sector I/Os run at 1/5th
the speed they would with a good controller. Interestingly enough, that's the
fault of the controller, not the host, even though you'd think the cost of
doing programmed I/O to fill/empty the sector buffer is a big deal. At least
that's my conclusion from the fact that less than 4:1 interleaving is efficient
on both Pro 350 and 380, while 3:1 interleaving is much slower on both. If the
buffer fill were a factor you'd think that the 380, with its much faster CPU,
would allow the use of a smaller interleave factor.
That said, you'd think that DMA would make a 1:1 interleave controller much
more feasible. And Bjoren also mentioned Ethernet. The DECNA is not to
horrible without DMA because you can use its on-board memory directly as host
packet buffers, though CT bus based memory is as I recall slower than
motherboard memory. Still, one wonders why they didn't use a correctly
designed Ethernet chip like LANCE, either with local memory or with DMA.
paul
