End-to-End Internet Packet Dynamics. Paxson, V. IEEE/ACM Transactions on Networking, 7(3):277-292, June, 1999.
End-to-End Internet Packet Dynamics [link]Paper  bibtex   
@article{ Paxson99,
  author = {Vern Paxson},
  title = {End-to-End Internet Packet Dynamics},
  journal = {IEEE/ACM Transactions on Networking},
  year = {1999},
  volume = {7},
  number = {3},
  pages = {277-292},
  month = {June},
  annote = {This paper uses two set of traces from 94 and 95 to characterize the amount of packet loss, packet delay and out-of-order delivery that occurs. 100 Kbyte of TCP data was sent and traced at both the receiver and sender. The paper says that larger windows were used for the second set, but gives no figures. The method of measuring bottleneck bandwidth with packet pairs is discussed, and Packet Bunch Mode (PBM) is used to overcome problems of packet pair such as multi-channel links. The loss ratio average is reported to be 3% and 5% for each of the sets, with many connections lossless. The lossiest half of the connections lose on average 6% and 9%, respectively. The forward and reverse loss ration is relatively independent although in the same order of magnitude. Two mechanisms work here: reverse losses should be less since less data is sent, but higher since TCP does not adapt to ack loss, only data loss. Losses are typically bursty with a conditional probability of 20-50%. Redundant retransmissions, i.e where the data had/would arrive at the receiver where classified into three types: 1 unavoidable, due to large ack loss. 2 coarse feedback due to the go-back-n method used. 3 bad RTO. A majority of the redundant retransmissions were of type 2 for both sets, and hence SACK is argued for.},
  url = {papers/Paxson99_end_end_packet_dyn.ps.gz},
  submitter = {Johan Garcia},
  bibdate = {Monday, May 08, 2000 at 17:26:42 (MEST)}
}

Downloads: 0