[email protected] (Anne & Lynn Wheeler) writes: > a big issue with tcp throughput is slow-start as mechanism for > congestion control/avoidance ... aka in enormously large heterogeneous > network with dozens of hops end-to-end and bursty traffic ... there is > relatively high probability of periodic congestion. dropping a packet > and then restarting slow-start ... can enormously cut throughput.
aka ... doesn't distinquish between missing packet because of transmission error and dropped packet because of congestion. it is possible to show that dynamic adaptive rate-based pacing can be stable and maintain higher throughput rates than slow-start in the face of packet drops. for some topic drift ... in hsdt there were some fiber-links with 10**-9 bit-error-rate with 15/16s reed-solomon FEC ... which resulted in effective 10**-15 bit error rate ... approx. the same as ibm mainframe channels of the period. there was also a gimmick for selective resend (on packet drop) ... where instead of retransmission of the same packet .... would transmit the 1/2 rate Viterbi FEC ... the receiving end could have both the original packet in error and the Viterbi FEC packet in error (despite 15/16 reed-solomon FEC) ... and still be able to reconstitute the original data. if error rate continued at high enough level ... transition to sending the 1/2 rate Viterbi as part of original transmission ... cuts effective rate in half ... but during periods of extremely hostile transmission ... packets still get through. -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
