• linux kernel
• macchine virtuali
• qemu
• virtual machines

Network performance is a critical aspect in Virtual Machine systems and its
importance is becoming increasingly important in the world of computing.
These systems are commonly employed in the IT departments of several organizations, since they can help to
build services that are highly reliabile, availabile and secure, improve efficiency
in computing resource usage, and so on.
In this thesis we are going to analize the state of the art of virtual machine networking, evaluating advantages
and drawbacks of the existing solutions.
We then propose a new approach, showing that with a small amount of code modifications, we can bring a classic emulated
network device (we take \texttt{e1000} as a reference example) to performances that are similar to the performances of
paravirtualized solutions.
However, this is not enough to push the performance to the limit (expecially latency).
Therefore, we put together the lessons learned, and introduce a new minimal paravirtualized solution, that can be implemented
in total with about 2400 lines of code (driver part and board emulation part) and it is intended to outperform the currently
existing solutions.