[Info-vax] Poulson at hot-chips 2011
glen herrmannsfeldt
gah at ugcs.caltech.edu
Wed Aug 24 18:36:23 EDT 2011
Bob Koehler <koehler at eisner.nospam.encompasserve.org> wrote:
(snip)
> Instruction replay is essentially what a VAX uses when an instruction
> gets a page fault. I suspect all virtual memory systems do something
> like it.
IBM S/370 has instruction nullification, to guarantee that nothing
that matters has been changed, and the PSW still points to the
instruction. For S/360 interrupts, it was usual that the PSW
pointed to the next instruction, and the instruction length code
was needed to back up (in interrupt routines) to find the offending
instruction.
S/370 also has interruptible instructions, which set the registers
appropriate for restart.
> Recovery from predictibly bad data is how old 1/2 inch tape drives
> got 6250 BPI net by actually recording about 9000 bits per inch.
Well, that is a different problem.
800 BPI tapes use NRZI coding, where there is a flux transition
for a '1', and no transitions for a '0'. It written odd parity,
there is at least one transition for each character. If the head
azimuth is off, then the transitions might seem to belong to the
wrong character.
1600 BPI tapes use Phase Encoding (PE), which is pretty much what
floppy disks call FM, and we call single density. There is a flux
transition between characters, and one in the middle of the cell
or not, depending on '1' or '0'. The maximum flux transition
density is then 3200 per inch. The clock for each track can
be recovered separately, such that head azimuth is relatively
unimportant, and the ability to reliably read tapes from another
drive is much improved.
6250 CPI is GCR, Group Code Recording. At 9000 flux transitions
per inch, it is only about 1.5x the character rate, instead of
twice as for PE. A set of codes with appropriate combinations
of '1' and '0' is chosen to code for a group of data bits.
-- glen
More information about the Info-vax
mailing list