GGrantIndex
← Search

CSR/PDOS: Concurrent, Direct Network Access: High-performance, Low-overhead Network I/O Virtualization

$461,988FY2007CSENSF

William Marsh Rice University, Houston TX

Investigators

Abstract

The economics of supporting a growing number of Internet-based applications has created a demand for server consolidation, and thus a resurgence of interest in virtualization. Virtualization systems for commodity hardware virtualize processor, memory, and I/O devices in software. Although this enables these systems to support a wide range of hardware, it also leads to significant overheads. We have developed Concurrent, Direct Network Access (CDNA), a new I/O virtualization architecture combining software and hardware that reduces the overhead of network virtualization. This architecture provides untrusted virtual machines safe, direct access to the network interface. While CDNA dramatically improves the efficiency of I/O virtualization, a notable gap between native and virtualized I/O performance still exists. This project's objective is to eliminate this performance gap without sacrificing the generality and manageability of software-based I/O virtualization. This leads to three main thrusts. First, we are working to mitigate the remaining overheads of the CDNA I/O virtualization architecture, which include memory protection and the scheduling of interrupts. Second, we are developing mechanisms to support full virtualization, memory protection using IOMMU hardware, and protected DMA by a conventional NIC. This will improve the generality of the CDNA I/O virtualization architecture. Finally, we are developing mechanisms to enable migration among systems with and without CDNA and to provide mechanisms for network resource provisioning. This will facilitate system managament for virtualized servers using the CDNA I/O virtualization architecture. Altogether, this research will make CDNA a complete I/O virtualization solution that provides efficient, general, and manageable I/O virtualization.

View original record on NSF Award Search →