Technical Summary
This document describes a modified sender-side algorithm for managing
the TCP and SCTP retransmission timers that provides faster loss
recovery when there is a small amount of outstanding data for a
connection. The modification, RTO Restart (RTOR), allows the transport
to restart its retransmission timer so that the effective RTO becomes
more aggressive in situations where fast retransmit cannot be used.
This enables faster loss detection and recovery for connections that are
short-lived or application-limited.
Working Group Summary
It is the consensus of the TCPM working group to document this
alternative algorithm, given the potential performance benefit. The work
has mostly been driven by the authors, but the document has been
reviewed in detail by several experts and the content has been modified
accordingly. Performance experiments in simulations and testbeds have
been performed and published by the authors and the experimental results
have been reviewed in several TCPM meetings. At the time of writing,
there is only limited deployment experience.
Two issues have been discussed extensively in the working group. First,
any reduction of the retransmission timeout duration inherently comes
along with a risk of negative impact on TCP performance, e.g. in mobile
networks with highly variable RTT. The current understanding is that
this risk is low and that the algorithm is conservative and relatively
robust, but further experimentation has to confirm this. Second, the
Linux operation system uses the "Tail Loss Probe" method discussed in
Section 6, which is similar but more complex. This method was not
adopted in TCPM since it depends on FACK error recovery method, which
has not been standardizes so far.
Document Quality
This document was also last called in TSVWG, since it specifies an
algorithm that can be applied both to TCP and SCTP. As a result of WGLC
comments the applicability to SCTP has been better explained, including
the SCTP API. One issue is that TCP and SCTP use slightly different
terminology for comparable concepts. In order to keep the document
simple, it was decided not to add another, duplicated description of the
algorithm using SCTP terminology.
Personnel
The document shepherd is Michael Scharf <michael.scharf@alcatel-
lucent.com>. The responsible Area Director is Martin Stiemerling
<mls.ietf@gmail.com>.