On Fri, 30 Oct 2009 15:33:09 -0400 Alan Altmark said: >On Friday, 10/30/2009 at 05:14 EDT, "DUBOIS Laurent (EXT DUBOIS)" ><[email protected]> wrote: > >> I have a zTPF guy who tells me that z/TPF will support the enhanced DAT >> >> facility soon to improve performance. >> When I checked our CPU facility list >> installed, the machine native shows >> the enhanced DAT facility installed, but >> z/VM does not. >> >> I am wondering if this facility is z/VM generation option. >> >> Have anybody an idea ? > >No, you cannot persuade z/VM to use enhanced DAT ("EDAT"). > >Another religious war in the offing. One of the ways EDAT improves >performance is by eliminating page tables (leaving more room in the cache >for Good Things) and speeding up the address translation process. All >"pages" are 1MB (the size of a segment). When paging, that means you have >to swap 256 4K pages at once, and they have to be contiguous in real >memory. So while you get some good benefits, memory fragmentation can >create significant pain. > >And in a virtual environment, what if the guest isn't using EDAT? You >could end up with both 1M and 4K pages in the system, creating the need >for *more* memory management. <gaack> Intellectually, the best result >will be when the hypervisor and the virtual machines both use 1M pages. > >We have it on our to-do list, but it isn't a trivial piece of work. I >just disassembled/reassembled my wife's laptop, so the imagery is fresh: >The concept of a 4K page is part of the CP backplane and >zero-insertion-force sockets were not used. ("Who would ever want a page >> 4K?") It's soldered tight, as it were, so disengaging the rest of CP >from that infrastructure without damaging delicate components will be.... >challenging. :-) It's a good thing we thrive on such challenges!!
So are you saying it's just like a Mac? /ahw > >Alan Altmark >z/VM Development >IBM Endicott
