60 likes | 173 Views
RDDP TCP Mapping: Rough Consensus Summary. David L. Black RDDP WG Chair IETF Vienna – July 2003. Status of TCP mapping issues. “What a long strange trip it’s been” The Grateful Dead Most major TCP mapping issues are now closed Sender Alignment Markers CRC Summary of resulting positions
E N D
RDDP TCP Mapping:Rough Consensus Summary David L. Black RDDP WG Chair IETF Vienna – July 2003
Status of TCP mapping issues • “What a long strange trip it’s been” • The Grateful Dead • Most major TCP mapping issues are now closed • Sender Alignment • Markers • CRC • Summary of resulting positions • Clarification questions are ok, BUT ... • ... please do not re-open these issues
Sender Alignment • Alignment of upper layer PDUs (ULPDUs) to transmitted TCP segments • Resolution (San Francisco – March 2003) • Alignment is an optimization (MAY) • No MUST or SHOULD alignment requirement • Ok to Describe circumstances where sender use of alignment improves performance • Ok to use lower-case “should” to recommend alignment in those circumstances
Markers • Markers locate ULPDU boundaries in out-of-order reception cases (e.g., drops) • Resolution (mailing list – May 8, 2003) • Marker usage is OPTIONAL • Marker usage is determined by receiver • Senders MUST support markers AND no-markers • Receivers MUST support markers OR no-markers • Receiver in each direction determines what is used • The inter-marker interval is 512 bytes
CRC • Use of CRC to protect header and data • Resolution (mailing list – July 9, 2003) • (A) MUST implement CRC support • (B) CRC usage is all-or-nothing on header + data • Note: One vs. two CRC issue is still open • (C) CRC usage is bidirectional: both or neither • (D) CRC omission requires consent of both ends of connection – i.e., either end can force CRC usage • (E) CRC usage is OPTIONAL [but, c.f. (D)] • (E’) CRC usage is admin. config, not visible to protocols • ULP design discussion in applicability statement draft
Last? CRC issue • (B’) One vs. Two (header and data) CRCs? • Case for Two: • Fast placement: check header, then can immediately place data • Don’t have to wait for data in order to validate header • Case for One: • Single CRC not an obstacle to flow-through design – data size limited by packet size • Simpler to generate and check one CRC • Header/data boundary = layering abuse