Abstract

We characterize an inefficiency in the current specification of SCTP’s congestion control, which degrades performance (more than necessary to be “TCP-friendly”) when there are multiple packet losses in a single window. We present an SCTP variant, called New-Reno SCTP, which introduces three modifications. First, a Fast Recovery mechanism, similar to that of New-Reno TCP, is included to avoid multiple congestion window (cwnd) reductions in a single round-trip time. Second, we introduce a new policy which restricts the cwnd from being increased during Fast Recovery, thus ensuring that the newly introduced Fast Recovery mechanism maintains conservative behavior. Third, we modify SCTP’s HTNA (Highest TSN Newly Acked) algorithm to ensure that Fast Retransmits are not unnecessarily delayed. We show that New-Reno SCTP performs better, while at the same time, still conforms to AIMD principles. Also, we compare these two variants of SCTP with New-Reno TCP, SACK TCP, and FACK TCP under six different loss scenarios. Our results show that New-Reno SCTP is as robust as TCP FACK, and more robust than the others.

5 Figures and Tables

Cite this paper

@inproceedings{CaroCongestionCS, title={Congestion Control: SCTP vs TCP}, author={Armando L. Caro and Keyur C. Shah and Janardhan R. Iyengar and Paul D. Amer and Randall R. Stewart} }