Back in the days of uncompressed tape blocks (reel-to-reel, & early
3480) there was a nice one-to-one correspondence between a fixed point
on the tape within a block and all the bits in a byte in parallel tracks
on the tape, so it was actually relatively simple to pass the tape past
the heads in the reverse direction and read all the bytes within a block
in reverse order and all the blocks in reverse order at full read speed
with the tape continually moving in reverse.  Tape sorts made use of
that to eliminate the repeated several-minute rewind time each time
another pass over the tape data was required.

I never had to deal with  reverse-read I/O, but PofOp indicates that one
supplied the address of the high end of the buffer as the channel
program data address and the channel decremented rather than incremented
byte addresses while transferring bytes to memory, so that the bytes of
a block would be restored to correct order in memory but would be
high-justified rather than low-justified in the buffer for short blocks.
 So the channel hardware itself effectively reversed the order of bytes
as they were read in reverse order.

I believe once the physical tape blocks became compressed (3480C and
later) it became impossible to actually read the physical blocks with
the tape moving in reverse and extract the original bytes on the fly.
Read reverse on such drives would no doubt have been emulated by the
device by backspacing over blocks, reading physical blocks with a
forward read, extracting the bytes, figuring out what bytes to supply to
the channel in reverse order to be consistent with what read-reverse
should have produced, then backspacing again to get to preceding block
and repeating.  Reversing the tape direction repeatedly while reading an
entire tape in reverse would  totally kill performance on one of those
drives.

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN

Reply via email to