|
本帖最后由 nfsking2 于 2011-12-11 20:57 编辑
koshuen 发表于 2011-12-7 20:26 ![]()
乍一看说的很有道理,但是百度了一下:
VT-x/VT-i:主要在处理器中实现,允许虚拟机直接执行某些指令,减 ...
虚拟化技术是个很笼统的东西,无论宿主机是32位还是64位,要跑64位虚拟系统,Intel CPU的唯一要求就是VT-x,跟VT-d无关。
VT-d是为了实现DMA隔离,或者说DMA Remapping。这个东西在负载比较高(比如同时运行多个虚拟机)的宿主系统上确实很有效,因为这个时候一个硬件要同时响应不同虚拟机的多个请求。不过普通人就跑一两个虚拟机,几乎对性能毫无影响。VT-d还需要硬件的配合,并且多数时候是针对PCI-E设备的(不止是磁盘读写才叫I/O),而不是光CPU和虚拟机软件支持就行的。
While VT-d enables the direct assignment of devices to guest VMs, it does not directly facilitate the efficient sharing of devices across multiple guest VMs. Such efficient sharing is not feasible without fundamental changes in the way that devices present their resources to the platform. Further work is being done in the PCI-SIG [21] [22] to enhance the PCI Express* specifications to enable devices to be shared.
Briefly, these extensions enable PCI Express devices to support multiple virtual functions, each of which can be discovered, configured, and managed. This allows the direct assignment of a virtual function to a VM using VT-d, thus allowing a single physical device to be sharable among multiple VMs.
The importance and applicability of these sharable PCI Express devices may be largely dependent upon the performance requirements, usage model, and market segment in which they may be deployed.
|
|