Summary: | TCP does not perform well in networks with high packet error rate like those with wireless
links because TCP assumes network congestion to be the major cause for packet losses. Wireless
losses make TCP unnecessarily initiate congestion control mechanism which results in poor
performance in the form of low throughput and high interactive delay. The link layer scheme,
which employs data link protocols in the base station and mobile host to retransmit lost packets
over wireless link, may be employed to hide wireless losses from TCP, but the problem of
competing retransmissions between TCP and link layer may occur, causing unnecessary duplications
and significant degradation in TCP performance.
This thesis investigates, through computer simulations, the end-to-end effects of link layer
retransmissions over a low data-rate wireless link on TCP Reno. The results show that, by using
the more effective selective-reject ARQ in the link layer, the problem of competitive retransmissions
between TCP and link layer is much less serious than previously reported. It is also found
that a non-sequencing link layer in combination with fragmentations of datagrams at the base
station and mobile host can be employed without significantly degrading TCP performance, thus
avoiding re-sequencing buffers and complex logic for handling out-of-sequence packets that
would otherwise be needed for a sequencing link layer protocol. The link layer modifications for
best-effort retransmissions, with a suitable division of the wireless-loss recovery function between
TCP and link layer, are proposed to reduce the possible adverse effects of link layer resets and
increased round trip time estimates from link layer recovery on TCP. The effects of link layer
reliability, controlled by the maximum number of link layer retransmissions or the maximum link
layer recovery time, on TCP throughput and round trip time estimations are studied. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|