Review: Machine-Independent VM...

From: Steve Arnold (stevearn_at_microsoft.com)
Date: Wed Feb 11 2004 - 14:22:26 PST

  • Next message: Nathan Dire: "Mach VM Review"

    The main idea behind this paper is that they have created a VM system
    which separates the hardware from the software and thus can be easily
    ported. There is still a small lower layer that must understand the
    hardware, called the pmap. However after that, everything is the same on
    all platforms.

     

    Memory objects (as they call them) are managed in a doubly-linked list.
    This help to support sparse VM space, rather than have huge, but mostly
    empty, tables. They also keep a memory allocation queue for managing
    free pages.

     

    One of the major features of the VM system is to support sharing. It
    supports Copy-on-write. When a copy-on-write is performed, the memory is
    not actually copied, until it needs to be written by one task or
    another. It does this by maintaining shadows. They also mention that the
    page can be written by the user.

     

    In their performance comparisons, it runs slightly better than 4.3bsd.
    Of course, 4.3bsd didn't have that great of VM support in the first
    place. One might wonder if you were to write a hard-ware specific
    system, that you might get much better performance (although not as
    portable). Also it seems like we have actually moved more toward
    standardization of hardware, so is this much of an issue anymore?


  • Next message: Nathan Dire: "Mach VM Review"

    This archive was generated by hypermail 2.1.6 : Wed Feb 11 2004 - 14:23:31 PST