We have a variety of outstanding issues regarding transport and ICE. This email tries to summarize the approach adopted in -07 for these issues.
* ICE-TCP is currently MTI and thus a normative dependency. However, it's unclear when ICE-TCP will be finished. * The draft needs to specify framing and a timeout for TCP or other stream-oriented protocols that do not provide such notifications themselves. * We have the AIMD proposal in an appendix, * The authors promised to merge [App]Attach and [App]AttachLite into [App]Attach, but have not done so. * Feedback about various related issues from WGLC. To reconcile these issues, the following changes have been made. * New text has been inserted to clarify that new ICE transport protocols and extensions (such as ICE-TCP) can be used within RELOAD by defining new codepoints. * References to ICE-TCP have been removed from the specification. * "Reliability for Unreliable Links" is renamed "Simple Reliability (SR)". * The header from SR is now used for TCP as well. It wastes a few bytes, but allows the sender to receive ACKs from the receiver, required to time RTT and determine when the link should be regarded as dead. * ICE Transports are now labeled Overlay Link protocols, as they specify not only transport but also framing, etc., eg DTLS/UDP with SR * *AttachLite are removed. In their place, new overlay link protocols are defined. DTLS/UDP with SR no ICE and TLS/TCP with FS no ICE. Text is added that defines how lite and full ICE implementations interoperate. (text below) * The AIMD proposal is deleted in its entirety. In its place is a reference to future definitions of semi-reliable datagram-oriented protocols over UDP such as SCTP and DCCP currently under discussion in TSV. Preference is recommended for these, as any would be superior to all current transport options (high-performance without head-of-line blocking). _______________________________________________ P2PSIP mailing list [email protected] https://www.ietf.org/mailman/listinfo/p2psip
