On Thu, 19 Jan 2006 10:54:16 -0600, Craddock, Chris wrote: >> >zAAP engines are a slightly different specialization. They are not >> >visible in the normal scheme of things and they can't field (I/O) >> >interrupts. > >> Is this "they can't" because of the contents of control register 6, or >> because of something else? > >I am not really sure. At some level the I/O interrupt mask has to be >off, but the key question is where is that set? Don Deese said that the >personality of the cpu is set in the HSA by PR/SM as part of LPAR >dispatch. > >That has the ring of truth about it because among other things, it >wouldn't require the overt cooperation of the operating system. z/OS and >zVM are both more aware of their environment than zLinux - which >basically doesn't care what the engine is.
The I/O interrupt issue has me wondering though -- are folks claiming that a zAAP cannot process paging I/O for itself? Surely the JVM is not page- fixed in storage, is it? (I don't see obvious evidence of that on our systems, but they ARE the size of small continents so that might be part of the reason why, I suppose.) I am expecting that the zAAP's control register 6 has a sufficiently 'different' I/O interrupt subclass value to allow for paging I/O and not for "normal" I/O, thus implementing the spirit of the zAAP nicely. >> Why do you say that the overhead is greater to dispatch a zAAP than a >> non-zAAP CPU? (I'm curious.) > >Well for one thing, when the dispatcher is running on a normal CPU it is >already running on the engine where the application code will be >dispatched. There is a separate (not externally documented) signaling >protocol to get work fired up on a zAAP. Even on its best day that is >going to entail more path length than an ordinary dispatch. At least one of the (mumble-multiple) zAAPs here is usually active with mumble-multiple JVMs having available work at most any given point when the sun is visible in the sky here. Often it is more than one zAAP busy. I don't see the non-zAAP working as hard as the zAAPs a lot of the time during the day. Given that environment I don't see why the dispatcher wouldn't dispatch an SRB... especially one involved in fielding page I/O, for example. (But I don't recall having seen one on a zAAP for certain yet, either; I suppose I can go dig through some dumps.) -- Tom Schmidt Madison, WI ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
