Date Added: Nov 2009
This paper describes a series of optimizations the authors performed on the Xen network IO subsystem. The original Xen virtualization architecture was introduced in. The idea of separate I/O domains was described in. This work was able to achieve significant performance gains in both transmit and receive throughput by employing a range of hardware and software optimizations. Most of the optimizations revolve around sending larger chunks of data between the driver domain and the guest domains. This is achieved by deferring the TCP segmentation of data until after the inter-domain crossing and using superpages (4MB Page Sizes).