Kim discusses a simulation technique that uses network-based calculus with packet-level generation for the simulation of transmission control protocol (TCP) networks. He addresses the tradeoffs of this hybrid simulation scheme, namely, the loss of some simulation detail for the gain of simulation speed.
The work presents solid arguments, and provides a simulation comparing the performance of the hybrid simulation to the performance of the packet-level simulation. However, the paper would have been more complete if some other issues were addressed. First, there was no simulation with different packet sizes: the packet size was fixed for most experiments up to 1,000 bytes, but traffic characteristics show high packet size variability. Second, there was no simulation with real foreground traffic, using previously captured data. I suspect that statistical laws (for example, the law of large numbers) will make the results converge, but addressing these issues could make for an interesting set of changes.