What is the best paper about? The researchers present VMSimInt, a new approach for packet-level simulations based on the integration of Virtual Machines (VMs) into the simulation environment. The framework in the study is based on an event-driven simulation program, which models the network and is controlled by a simulation calendar. There are three parts in the framework: an extended QFMU, a QEMU Adapter class in the simulation program, and a relay program.
There are several important implementation aspects. For instance, VMSimInt is based on QEMU as it is Open Source software and its virtual components are split into a virtual device and a device backend. Another aspect is that QEMU is extended to clock control and I/O direction. When it comes to clock control, the two sources of time are differentiated: wall clock and simulated time.
How do researchers go about the evaluation of VMSimInt? By dividing it into three parts: 1) comparison of the approach to the ns-3 simulator; 2) comparison the behavior of different Linux kernel versions; and 3) evaluation of VMSimInt runtime performance. The results are intriguing. The first part shows that VMSimInt provides results matching those of NSC + ns-3. In the second comparison phase, the kernels with an initial congestion window of 10 packets behave identically and start more quickly than the unmodified 2.6.26 kernel. The last phase displays that the runtime increases linearly for the NSC simulator with increasing number of stacks.
The real contribution of the paper is that it demonstrates that the simulations performed with VMSimInt provide results comparable to NSC + ns-3 approach, which is widely used and the basis of recent research in the TCP area. It also shows that the computation effort and memory consumption are higher than with existing tools, but are still viable for typical TCP simulations.
Last, but not least, the researchers are available for further information: VMSimInt source codes can be requested by e-mail.
You can read the full paper here