A commodity I/O device has no support for virtualization. A VMM can assign such a device to a single guest with direct, fast, but insecure access by the guest's native device driver. Alternatively, the VMMcan build virtual devices on top of the physical device, allowing it to be multiplexed across VMs, but with lower performance. We propose a technique that provides an intermediate option. In virtual passthrough I/O (VPIO), the guest interacts directly with the physical device most of the time, achieving high performance, as in passthrough I/O. Additionally, the guest/device interactions drive a model that in turn identifies (1) when the physical device can be handed off to another VM, and (2) if the guest programs the device to behave illegitimately. In this paper, we describe the VPIO model, and present preliminary results in using it to support two commodity network cards within the Palacios VMM we are building. We believe that an appropriate model for an I/O device could be produced by the hardware vendor as part of the design, implementation, and testing process.
This paper describes the design of a system to enable research, development, and testing of new software stacks and hardware features for future high-end computing systems. Motivating uses include both small-scale research and development on simulated individual nodes of proposed high-performance computing systems, and large scaling studies that emulate a sizeable fraction of a future supercomputing system. The proposed architecture combines system virtualization, architectural simulation, time dilation, and slack simulation to provide scalable emulation of hypothetical systems. Virtualization-based full-system measurement and monitoring tools are also included to aid in using the proposed system for co-design of high-performance computing system software and architectural features for future systems. Finally, this paper provides a description of the implementation strategy and status of the system.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.