A simulation study of differential fair buffer allocation

The guaranteed frame rate service (GFR) is viewed as the most promising ATM service for carrying TCP/IP traffic over large distances. We study the behavior of differential fair buffer allocation (DFBA) in the context of an ATM WAN carrying TCP Reno traffic, with a special focus on fairness issues. The factors investigated are the buffer size, the MCR allocation and other DFBA parameters, the presence of high priority rt-VBR traffic in the switch, and the round-trip-times experienced by the TCP sources. Along with several other results, our study shows that the DFBA proposal fails to provide a fair share of the buffer to VCs with different minimum cell rate (MCR) allocations and/or average RTTs. As a partial solution, we propose the use of weighted averages for calculating the drop probability.