On Sat, Jan 7, 2023 at 11:17 AM Paul Gilmartin < [email protected]> wrote:
> On Fri, 6 Jan 2023 15:33:44 -0600, Hobart Spitz wrote: > > > > > https://docs.google.com/presentation/d/1C7htK9I4rL-4pUZcdhfeA4-IY2Kjxm-HJQ4fuljMs3I/edit?usp=sharing > > ... > >Would you rather pass data in move mode (*nix piping) or locate mode > >(Pipes) ... > > > How does locate mode deal with a FANOUT stage? > Great question Paul. In Pipelines, the pointer to each record is sent to each output stream in turn. Each output stream sees the same buffer as input. If data is changed by a downstream stage, a new pointer to a new buffer is created. In zPipe, the plan is to duplicate the stack the required number of times. The effect is analogous, but different: One copy of each data record in the heap is pointed to two or more times by entries in different stacks. Each stack can be processed i independently, when it reaches the top. If data is changed, new space in the heap is used, and a new pointer to it, created. I have a zPipe implementation of LOOKUP that reads the top stack into a stem as detail records, does a delstack, and reads the next stack as a master reference. Then the lookup process proceeds. I hope this helps. > > -- > 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
