CSR: Small: I/O Virtualization at the Device File Boundary and its Applications

William Marsh Rice University, Houston TX

Investigators

Abstract

Virtualization allows multiple virtual computers to securely share the hardware resources of a physical computer. This has become a critical technology for security, resource consolidation and management in modern computers. Among hardware resources, input and output (I/O) devices have proven to be hard to virtualize because of the great diversity in their functionality and implementation; yet, I/O devices are very important for modern computers from smartphones to data center servers. They include not only human input and output devices such as sensors and actuators, but also computational resources like graphics processing unit (GPU) and field-programmable gate array (FPGA). Known I/O virtualization solutions often struggle between development effort and supported device functionality. This project aims at not only solving this impasse but also extending I/O virtualization from locally connected I/O devices to network I/O virtualization - I/O devices from computers available via network. It will also demonstrate the novel, compelling application cases of network I/O virtualization for both mobile devices and data centers. Device file, an abstraction used by Unix-like operating systems for I/O devices, can be the sweet spot as the boundary for I/O virtualization. The project's focus centers around three technical contributions: First, high-performance cross-computer file operations that allow file operations issued by one computer to be efficiently executed by device drivers residing in another computer, virtual or physical, even over network. Second, this project will provide fault and data isolation between multiple computers sharing an I/O device and provide correctness and performance guarantees to them without sacrificing the potential benefits of the device file boundary. Finally, this project will demonstrate the novel applications of network I/O virtualization for both mobile and server systems, including accelerator consolidation in data centers and I/O sharing between mobile devices. The project will leverage collaborations with industry leaders to ensure a timely transfer of technologies into industry and there is a broad impact on the commercial development of mobile devices and data centers. The project will provide new system components for teaching mobile and embedded systems to a diverse community of learners. Finally, as mobile devices have penetrated all walks of life and data centers have become the backbone of personal computing, the project has the potential to impact this global society and its economy by reducing the operational cost of data centers and by enabling novel use of mobile devices.