On 31 December 2013 13:23, Paul Gilmartin <[email protected]> wrote: > Most (?) of the complaints about (non-)shared areas stem from the > non-propagation of DDNAMEs through fork(). Ain't gonna get better > (NVFL, anyway). Because of ENQ conflicts between parent and child. > Extend the ENQ scope to job rather than address space? NVFL, again. > > I could imagine: > > o A scheme where an allocation could be transferred from parent to > child, freeing the ENQ in parent. > > o A server-client model in which the actual I/O is performed by the > parent that performs the allocation, and the data passed to the > child via sockets or POSIX pipes. Sort of like NFS, but it needs to > be better than NFS.
The likely problem is that - unlike the MVS subtasking model - in UNIX the parent process can go away leaving the child running. Would you leave the parent running just to handle the I/O work, or clean it up and transfer the ENQs and/or allocations to the child at that point, or just say Don't Do That, or...? Tony H. ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
