Powerpc - Context Switch.
  • Butterfly
    Butterfly
    rebraist
    Posts: 96 from 2011/4/6
    From: Naples, Italy
    Hi!
    I'm studying a bit morphos and I'm really amazed discovering what's in it.
    It's really beyond being "simply" an Amiga os like Operating System.
    I'm arrived at an inner look to Quark and the fact it's a pseudo L4 microkernel.
    You know, one of the issues of mukernel is context switch.
    Due to Ipc, kernel moves between user-kernel mode just often.
    Now one of the factors that worse performance context switch times is the number of cpu registers: higher it is, worse will be performances.
    Ppc has 32 general registers, and sometimes 32 floating point registers.
    In an internet article i read that morphos is an example of speed (the argument was microkernels) because of powerpc registers, so (wrote the author) you never will go on the ram stack, saving a lot of time.
    Two things are mutually exclusive: or register increase performance or register decrease performance.
    What of two?
    Thx!!!


    edit: Clearly stripping away lack of mmu times in a non protected memory environment.

    [ Edited by rebraist 26.11.2013 - 16:53 ]
    Mac Mini g4 1,5 mos 3.1 registered
    Powerbook g4 1,67 mos 3.1 unregistered
    Sam440 Os4.1.6
    Aros-Aros-Aros.
  • »26.11.13 - 15:40
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12200 from 2003/5/22
    From: Germany
    > i read that morphos is an example of speed (the argument was microkernels) because
    > of powerpc registers, so (wrote the author) you never will go on the ram stack, saving
    > a lot of time. Two things are mutually exclusive: or register increase performance or
    > register decrease performance. What of two?
    > edit: Clearly stripping away lack of mmu times in a non protected memory environment.

    Some speculation by an AROS/ARIX kernel guy:

    http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=38299&forum=14&start=180#720652
  • »26.11.13 - 16:17
    Profile
  • Butterfly
    Butterfly
    rebraist
    Posts: 96 from 2011/4/6
    From: Naples, Italy
    Yes, he talks about the overhead of mmu in context switch and, consequently, in an amigoid system (without paging) there's no such overhead, with better performances.
    Other things, i think are related to sync/async ipc and the fact that being abox a process of quark (the only process indeed) there's no need to save mmu in pcb.
    But still I don't understand why context switching with powerpc (and not sw related) should be better than x86 counterpart...
    Thk you!!!
    Mac Mini g4 1,5 mos 3.1 registered
    Powerbook g4 1,67 mos 3.1 unregistered
    Sam440 Os4.1.6
    Aros-Aros-Aros.
  • »26.11.13 - 19:24
    Profile
  • Visitor
  • Phil Wilson (Andrew Wilson) can finish the last main operating formal (John Riccitiello) in Objective of this season the starting staying by the resignation , because the new go of http://lol.vipmmobank.com/
  • »27.11.13 - 04:47