How good is Xen for simulating distributed applications?

Simulating large networks has been an important challenge we face in distributed systems research. A good solution for this however has been much sought after. People have found easy solutions in network simulation tools like NS2. Though NS2 is capable of simulating a large class of net-works with varied protocols, it has two major limitations. First, it provides a simulation of real network which in-evitably brings some bugs and limitations along with it. Secondly, it is a single process simulator which essentially means that it simulates all the networking components se-quentially. A real network however, is essentially parallel. Thus, NS2 does not in essence represent the true character-istics of a real network. There exists other alternatives to software simulators however. The most important among these is Emulab. However, Emulab has its own limitations as well. With Emulab,we are not able to use more than a few hundred nodes. Further, Emulab is a shared resource meaning that many users simultaneously use this single re-source. Thus, scalability is an important issue concerning with Emulab. With virtual machines gaining in popularity and with the overall good performance claimed by the Xen virtual machine monitor using para virtualization, it is the right time that we ask ourselves the question of whether we can use virtual machine monitors like Xen to achieve the scalability and performance required in distributed appli-cations. Can Xen provide us with the solution that we so desperately look for? Can it provide both scalability and performance as can be obtained if we were to use the real network resources? This paper seeks to answer some of these important questions in this direction.