The previous section showed that targeted optimizations can reduce the CPU overhead due to virtualization to the point where performance becomes I/O bound. This section describes strategies for further improving I/O performance and decreasing CPU utilization. The major areas for optimization include i) reducing CPU and interrupt controller virtualization overheads, ii) modifying the guest OS and/or its drivers, iii) modifying the host OS, and iv) accessing the native hardware directly from the virtual machine monitor. The last two techniques are departures from a pure hosted virtual machine architecture. Recall that the hosted architecture is designed with the requirement that existing host operating systems continue to run as usual, and that the virtualization software uses the host OS's API to access hardware devices.