5 Oct
2020
5 Oct
'20
6:10 a.m.
On Oct 5, 2020, at 12:30 AM, Chuck Guzis via cctalk
<cctalk at classiccmp.org> wrote:
On 10/4/20 8:58 PM, jim stephens via cctalk wrote:
At least on NOS, that's not how long block handling works. It uses 1MT (the main tape
driver) with 1LT as a helper in a separate PP. Each reads a portion of the long block
into its PP memory, hand off to the other PP to continue reading while in parallel the
first PP transfers the piece it read into central memory. This kind of ping-pong
algorithm is needed because the block transfer operations (both I/O and memory copy) are
blocking. They can be split up into individual word transfers but that isn't how the
problem is handled in this case, and possibly that wouldn't be fast enough.
paul
I had not seen your earlier no tape gap mentions.
The old CDC 6000 SCOPE 1LT driver. Since SCOPE user programs use
circular buffering, The PP overlay 1LT simply emptied the CM buffer and
wrote it, 12-bit word by 12-bit word to the drive controller, so long as
the CPU program kept the buffer filled.