Jones Optimality and Hardware Virtualization

Short Description
The growing popularity of hardware virtualization (VMware and Xen being two prominent implementations) leads us to examine the common ground between this yet-again vibrant technology and partial evaluation. A virtual machine executes on host hardware and presents to its guest program a replica of that host environment, complete with CPU, memory, and I/O devices. A virtual machine can be seen as a self-interpreter. A program specializer is considered Jones-optimal if it is capable of removing a layer of interpretational overhead. We propose a formulation of Jones optimality which coincides with a well-known virtualization efficiency criterion.

Website: www.cl.cam.ac.uk | Filesize: 190kb
No of Page(s): 7

Content
A program specializer is considered Jones-optimal if it is capable of removing a layer of interpretational overhead. We propose a formulation of Jones optimality which coincides with a well-known virtualization efficiency criterion.
A fully abstract programming language translation (an idea put forward by Abadi) is one that preserves program equivalences. We may translate a program by specializing a self-interpreter with respect to it. We argue that full abstraction for such translations captures the notion of transparency (whether or not a program can determine if it is running on a virtual machine) in virtual machine folklore.
…
Over the last few years, many academic and commercial hardware virtualization offerings have emerged (VMware [1] and Xen [5] are two prominent examples). A virtual machine monitor (VMM) is responsible for sharing out the hardware resources of the host system between several simultaneously running virtual machines (VMs). Each VM presents to its guest program a replica (with possible variations in the number and types of available I/O devices, amount of memory, etc.) of the host environment. In their seminal 1974 paper, Popek and Goldberg [20] described three requirements that a VMM must meet: efficiency, equivalence, and resource control. A VMM must not impose undue overhead, be faithful to the original hardware, and retain control over certain aspects of guest program execution (access to I/O devices, for example). Since we are not concerned with issues of concurrency, in the remainder of this article we shall assume that only a single VM is running and use “VM” in preference to “VMM” henceforth.
…

Get the file Download here