Peter H's version is a first-order match for how the history was related
to me when I first took over ownership of the "old" (pre-WLM) sysevents
approximately two and a half bazillion years ago.
To add a bit:

- It's the caller's intent about the duration of non-swappability that
  distinguishes which of the two you call.  If it's "just for a second
  or three" dontswap is fine; if it's for hours, or more usually
  (until the address space terminates, which might be months), you
  want transwap because of the preferred storage consequence.

- Historically, the latency costs were drastically different.  The
  differences narrowed, when states like "logically swapped" were
  added.  Transwap then was overkill for cases that dontswap could
  handle.

- Similarly, the consequences (how long might it take to vary a
  storage element offline?) vary drastically.  Waiting for an I/O
  or 3 to complete (the intended dontswap case) might be tolerable,
  waiting until some PC target address space (typically a long-running
  server) terminates ... what could happen if it used dontswap
  instead of transwap and it page-fixes non-preferred frames in the
  element you want offline... was "frowned upon".

- As others have noted, being non-swappable is NOT equivalent to
  page-fixing frames.  Work units running in non-swappable
  address space(s) and referencing pageable storage CAN take page
  faults (unless they're in state that does not permit them to be
  suspended, in which case they're ABENDed).

John Arwe

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to