One would certainly hope that "copying" an address space for purposes of
fork() would not require paging it all in.   That would be horrible, and I
seriously doubt that it works that way.

On Fri, Aug 23, 2019 at 12:26 PM Paul Gilmartin <
[email protected]> wrote:

> On Fri, 23 Aug 2019 10:41:55 -0500, Kirk Wolf <[email protected]>
> wrote:
>
> >I agree!
> >
> >FWIW - debugging performance of forked Unix process startup/overhead is a
> >mess - we had one customer who was seeing terrible performance when
> >fork/execing tiny little shell processes that did practically nothing.
>  It
> >was only happening on one of their LPARs.   The IBM Support center was
> >involved for a really long time, and I don't know if they ever really
> >figured out what was wrong.
> >
> Indeed.  Ouch!
>
> But I wonder how this plays with deferred loading of program objects
> (I forget the proper term).  Suppose a parent whose execution image
> is only partially loaded forks a child which accesses pages never loaded
> in the parent ...?
>
> So every page in the parent's execution image must be brought into
> storage so it can be copied to the child.
>
> >On Fri, Aug 23, 2019 at 8:24 AM Jerry Callen wrote:
> >
> >> According to the "Summary of changes for z/OS UNIX System Services
> >> Planning for Version 2 Release 4 (V2R4)" (see
> >>
> https://www.ibm.com/support/knowledgecenter/SSLTBW_2.4.0/com.ibm.zos.v2r4.bpxb200/bpxb2soc24.htm
> >> copy on write for fork() has been removed from z/OS. It will be
> interesting
> >> to benchmark the performance of fork() between 2.3 and 2.4 on a variety
> of
> >> processes.
> >>
> >> I wonder why this was done? (Hi, Peter Relson -- hint, hint...)
>
> -- gil
>
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions,
> send email to [email protected] with the message: INFO IBM-MAIN
>

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

Reply via email to