> > 1) I'm not seeing any CLCK/EXT/DSP entries for the logical > processor within the the trace. > > 2) The Physical CPU changes on every trace entry for the 15 second > period, There are only 11 trace entries.
It may be unusual to have 15 seconds of trace from all processors unless the system is mostly idle. In your analysis, are you considering trace entries which preceed this message? ******** TRACE DATA IS NOT AVAILABLE FROM ALL PROCESSORS BEFORE THIS TIME. You should generally not consider entries which preceed this message unless you are aware of the implications.i > > So it looks like the LCP is jumping all over the place onto > different PCPs without any trace indication, other than the CP number. > > > > From what Tom says z/OS doesn't know about the interrupt so the > LCP info is stored somewhere off z/OS and anything on that LCP would > remain undispatched until that LCP is put back on a PCP, which may > be the same one or a different one. > I would like to add that I believe there is interaction between z/OS > and PR/SM in state saving. Somehow somewhere. I believe I read > somewhere that with Hiperdispatch the MVS dispatcher works with PR/ > SM and has one WUQ (work unit q) per physical (or was that logical?) > cp. That change was made generally, regardless if Hiperdispatch is > used or not. z/OS is not directly involved in the undispatching of a logicial processor on which it is running, at least, not until the new EC12 processor. The EC12 can present a new "Warning Track" external interruption when it wants to undispatch a logical processor. This gives z/OS the opportunity to first undispatch the z/OS workunit so that z/OS can dispatch it on another logical processor. Jim Mulder z/OS System Test IBM Corp. Poughkeepsie, NY ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
