Review of Denali Isolation Kernel (By Whitaker et al)

From: Prasanna Kumar Jayapal (prasak_at_winse.microsoft.com)
Date: Mon Mar 01 2004 - 17:52:49 PST

  • Next message: Cliff Schmidt: "Review of Whitaker et al. "Scale and Performance in the Denali Isolation Kernel""

    This paper (Scale and Performance in the Denali Isolation Kernel) talks
    about the design and implementation of a virtualization layer that
    supports the execution of un-trusted internet services on a shared
    infrastructure.

    Denali emphasizes the configuration of one application per VM, and goes
    further by reducing the role and capability of the operating system that
    is sandwiched in-between. Standard operating systems cannot be run on
    top of Denali, only simplified sub-operating systems may run. In
    addition, the authors considered only independent applications that did
    not require sharing amongst each other. This is a primary difference
    when compared to the other VM paper, Disco that addresses sharing and
    resource management among different applications.

    The instruction set architecture of Denali is very similar to x86, but
    only contains a subset of it. It hides the non-virtualizable aspects for
    simplicity. However, it introduces couple new virtual instructions and
    some virtual registers. These registers expose system information and
    also serve as a sharing mechanism. Isolation kernel implements the
    virtual architecture and multiplexes physical resources across
    competeing VMs. It defines the size and the range of a VM, when it is
    instantiated. I/O devices has exposed as virtual devices which are
    similar in functionality to their physical counterparts but expose a
    simpler interface. CPU multiplexing consists of two policies - a "gate
    keeper policy" which enforces control, by choosing subset of active
    machines and the "scheduler policy" which controls context switching.

    The evaluation section was nice to read through and it kind of convinces
    that the performance overhead of virtualization is reasonable.

    Overall this was an interesting paper and I am really curious to see how
    well the web servers, search engines and the content management systems
    adopt to this technique.


  • Next message: Cliff Schmidt: "Review of Whitaker et al. "Scale and Performance in the Denali Isolation Kernel""

    This archive was generated by hypermail 2.1.6 : Mon Mar 01 2004 - 17:52:42 PST