Performance Optimization of PR-SCTP for Small Messages

• Stream Control Transmission Protocol (SCTP) was defined by the IETF Signaling Transport (SIGTRAN) working group to transport SS7 signaling traffic over the Internet. SCTP can also be used as a general purpose message based transport protocol. A partially reliable extension of SCTP, PR-SCTP has been considered as a candidate for prioritizing content sensitive traffic at the transport layer. By using PR-SCTP, it is possible to define various reliability policies for different application messages. An application chooses (re)transmission policy on a per message basis. A special control chunk called Forward Cumulative TSN (forward_tsn) is used to provide partial reliability. When a message is abandoned by PR-SCTP, for instance the associated lifetime is over, a forward_tsn chunk is sent to the receiver to tell it not to expect that particular message any more. PR-SCTP can offer a trade-off between timeliness and reliability for application messages when network resources become congested. Several applications such as streaming multimedia, IPTV transmission, and SIP signaling have been shown to benefit from this. Our previous work, however, suggests that the performance gain can be very limited in a network with competing traffic when message sizes are small. In this case, the most important factor is the inefficiency in the forward_tsn mechanism of PR-SCTP. In this work, we examine the forward_tsn inefficiency in detail and propose several solutions to overcome it. Moreover, we implement and evaluate one of the proposed solutions that takes advantage of the Non-Renegable Selective Acknowledgement (NR-SACK) mechanism available in the current FreeBSD OS. A receiver can use NR-SACKs to selectively acknowledge out-of-order but non-renegable data. Our results show a significant performance gain for PR-SCTP with NR-SACKs for small messages. In some scenarios the average message transfer delay is reduced by more than 75%.

Ämnesord

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Publikations- och innehållstyp

vet (ämneskategori)

kon (ämneskategori)