23 Apr
2020
23 Apr
'20
6:35 p.m.
Well I turned on block mode DMA on my MTI controller, booted RT11 5.5,
did an init on VM0: and tried to copy a .dsk file of 2500 blocks in size
with 2mb of PMI memory from an 11/84 (CA, bad memory for Q Bus).
First time was not good:
.copy dungeo.dsk vm:
Files copied:
?PIP-F-Input error DK:DUNGEO.DSK
.dir dungeo.dsk
DUNGEO.DSK 2500P
1 Files, 2500 Blocks
34488 Free blocks
Second time was better, took ~4 seconds.
.copy dungeo.dsk vm:
Files copied:
DK:DUNGEO.DSK to VM:DUNGEO.DSK
Switched off block mode and tried again:
Takes 5 seconds.
Changed burst length from 16 words to 2:
Took 25 seconds.
4: 23 seconds
16: 5 seconds
Turning off the throttle mode (which requires Block mode and 16 word
transfers) and it took 4 seconds. No speed increase.
Then I decided to try using 2mb of crummy Q bus memory:
16: 5 seconds
8: 15 seconds
2: 26 seconds
Didn't do 4.
So it looks like what really speeds things up is having a large burst of
Q busy transfers, going from 2 word transfers to 16 speeds things up by
a factor of five. Makes sense, controller is spending less time setting
up the transfer. Somewhat interesting that PMI doesn't really help, I
understand that the controller is talking to PMI at normal Q bus speeds,
but I would expect something for the CPU to manage the VM: driver.
Unless the CPU just says "Oh, VM:? Point your data right into memory and
be done with it".
Moral: PMI memory does not really speed up disk to memory performance.
Good to know. Anyone know a good benchmark to see if PMI is worth
anything in terms of performance? Maybe the unibus map interfacing
directly to PMI speeds things up a lot by allowing full hog mode DMA
(which in theory could mean that an 11/84 could run a RH11-C controller
and keep up with disks like the RM02 and RM05)
Thinking along those lines, was the 11/70's MASSBUS channels nothing
more than RH11-C's attached to the old FASTBUS on the 11/45 cpu core
(which is what an 11/70 really is, with cache) or did they port right to
the memory box?
C