New TCP Flavours - Simon Fraser University

Much better than the old TCP Flavours Rajon Bhuiyan 1 Why are the new flavours better? Efficiency! When TCP Tahoe and Reno were created, we did not know as much about the internet as we do now. New technology has been created that better suit the model of how the internet actually works. Specialization Not all machines connect to the internet under the same

circumstances. Some of the new TCP flavours provide enhanced performance when dealing with networks that are not the usual type. Integrability TCP New Reno, Vegas and Hybla can all play nice when connecting to networks currently running TCP Reno. Rajon Bhuiyan 2 Varieties TCP New Reno SACK Variant TCP Vegas

TCP Hybla TCP BIC, CUBIC, Compound TCP, Fast TCP, H-TCP, and many more Rajon Bhuiyan 3 TCP New Reno The most widely deployed of the new TCP flavours Has many phases in common with TCP Reno, like slow start, fast retransmit and additive increase-multiplicative decrease The main difference between plain Reno and New Reno is in

the fast recovery phase, which allows New Reno to deal with multiple packet losses In New Reno, the fast recovery phase lasts until all of the segments which were outstanding when it went into fast retransmit are accounted for. TCP New Reno can not determine what the second lost packet was until it receives an ACK for the first lost packet TCP SACK is an improvement over New Reno where segments are acknowledged selectively instead of cumulatively. Rajon Bhuiyan 4 TCP Vegas Developed by Lawrence Brakmo, Sean OMalley and Larry

Peterson. Unlike TCP Reno, Vegas uses the difference between expected traffic and actual traffic to calculate the size of the congestion window. Both increase and decrease of the rate is additive Because TCP Reno keeps increasing its sending rate until a packet is lost, it will always incur packet loss at some point or other. Vegas converges so it does not lose packets. Vegas can have 40 to 70% better throughput than TCP Reno with less than half the packet loss. Its disadvantages are that it can not efficiently deal with paths being rerouted and will also get an unfairly small share of the bandwidth when on a network that is running TCP Reno. Rajon Bhuiyan

5 TCP Hybla Protocol that is specially adapted to perform well in networks that contain satellites The problem with the usual TCP protocols is that the send window grows by one each time an acknowledgement is received, which severely penalized satellite and wireless devices on the network because they have very large round trip times TCP Hybla attempts to remedy the unfairness of the current TCP Flavours by removing the congestion windows dependency on the RTT

Instead of basing its growth on the RTT, TCP Hybla attempts to match the growth of its congestion window to the RTT0 of a reference connection Rajon Bhuiyan 6 Hybla Graph Rajon Bhuiyan 7 Conclusion The message you should all take away from this is that

internet technology is constantly evolving, and we are not currently using the best possible technology, just the best that was available back when the network was made. Rajon Bhuiyan 8 Bibliography Mo et al, Analysis and Comparison of TCP Reno and Vegas,. Available, http://netlab.caltech.edu/FAST/references/Mo_comparisonwith TCPReno.pdf, accessed March 30, 2008.

Anonymous, A comparative analysis of TCP Tahoe, Reno, New-Reno, SACK and Vegas. Available, http://inst.eecs.berkeley.edu/~ee122/fa05/projects/Project2/S ACKRENEVEGAS.pdf, accessed March 31, 2008 DEIS Department, Hybla home page. Available http://hybla.deis.unibo.it/, accessed March 31, 2008 Caini, Carlo and Firrincieli, Rosario, TCP Hybla: a TCP enhancement for heterogenous networks. Available http://www.cs.utk.edu/~dunigan/ipp05/hybla.pdf, accessed April 1, 2008 Rajon Bhuiyan 9