Throughput analysis of Non-Renegable Selective Acknowledgments (NR-SACKs) for SCTP

Preliminary work introduced Non-Renegable Selective Acknowledgments (NR-SACKs) and showed they (i) better utilize a data sender’s memory in both SCTP and CMT, and (ii) improve throughput in CMT. In this paper, we provide the latest specification of NR-SACKs, and extend the investigation of throughput improvements that NR-SACKs can provide. Using ns-2 simulation, for various loss conditions and bandwidth-delay combinations, we show that the throughput observed with NR-SACKs is at least equal and sometimes better than the throughput observed with SACKs. We introduce “region of gain” which defines for a given bandwidth, delay, and send buffer size combination, what range of loss rates results in significant throughput improvement when NR-SACKs are used instead of SACKs. In both SCTP and CMT, NR-SACKs provide greater throughput improvement as the send buffer size decreases, and as end-to-end delay decreases. Provided that the bandwidth-delay product (BDP) ⩾ send buffer size, additional bandwidth does not increase NR-SACKs’ throughput improvements for either SCTP or CMT. For BDPs < send buffer size, the throughput improvement decreases as the BDP decreases.