160 likes | 313 Views
QoS NSLP draft-ietf-nsis-qos-nslp-07.txt. Jukka Manner (ed), Georgios Karagiannis, Andrew McDonald , Sven van den Bosch. Resolved issues. Issue: Should there be an explicit ‘refresh’ indication? E.g. to prioritise existing reservations over new reservations
E N D
QoS NSLPdraft-ietf-nsis-qos-nslp-07.txt Jukka Manner (ed), Georgios Karagiannis, Andrew McDonald, Sven van den Bosch
Resolved issues • Issue: Should there be an explicit ‘refresh’ indication? • E.g. to prioritise existing reservations over new reservations • Resolution: No explicit indication added • Main problem is one of policing the indication • Issue: Should there be a separate message for TEAR (instead of flag in a RESERVE message)? • Resolution: Keep it as flag in a RESERVE message • Most of the processing is the same • Issue: What should happen when a ‘tear’ is received for an unknown reservation? • Resolution: Drop the message
Other updates • ‘QoS NSLP’/’RMF’ interactions • Conceptual model in figure 1 shows ‘QoS NSLP’ interacting with ‘NTLP’ and ‘RMF’ • Previous text suggested that the only thing that could pass between ‘QoS NSLP’ and ‘RMF’ was the QSpec • Clarified that the QSpec is not the only information relevant to admission control, policy control, packet classifier setup, etc • However, do not intend to define that interface (no ‘NSLP-RMF API’) • Multiple QSpecs • Previously allowed arbitrary stacking of QSpecs • Now limited to a stack of at most two • One end-to-end and one for local QoS model • Miscellaneous editorial work
Sequence number handling • Includes issues of node failure and restart handling • Discussion thread now on mailing list • Choice between circular and linear sequence space • Current suggestion: circular, wrapping as in RFC1982 • Addition of an epoc identifier
Security • Protocol components for authentication and authorisation • Add further discussion of use of GIMPS GIST channel security • Tokens • Need to work out what to specify • Reuse existing work? • E.g. POLICY_DATA from RSVP • Specify something new? • Leave it for now and specify as a later extension?
How do you perform receiver-initiated reservations? • Agree with application peer that it wants QoS reservation • Also an issue with RSVP • Agree with application peer who is going to initiate it • RSVP doesn’t have to worry about this • If receiver is initiating, need to be able to send NTLP messages upstream, so need to establish reverse path state • RSVP does this with PATH messages • Send RESERVE • RESV in RSVP
Alternative approaches? • Roland’s suggestion • Various issues due to end-to-end messages (not hop-by-hop) • GIST upstream query?
Fundamental question • How much do you rely on upper layer protocol interactions in the background?
Packet classifiers • What extra pieces are needed beyond the flow identification information in the MRI? • Which bits of the MRI can be ignored by the packet classifier? • A first attempt included in -07 • Provides flags to indicate which parts of the MRI are to be used for packet classification • Does not address all the issues of the Flow-ID draft • Is the current attempt correct? Is it sufficient? • How far should we go? • Are there fields we should add that we can’t get from the MRI? • However, there are issues of message routing and NAT to consider
Last node issues • Path extension
Last node issues • Path extension
Last node issues • Path extension • Path truncation
Last node issues • Path extension • Path truncation Rerouting description needs expanding
Error handling • Suggestion at Interim: • Share error classes between NSLPs • E.g. Information, Protocol Error • Share error codes between NSLPs • E.g. bad object length, resources not available • Then have NSLP-specific information • E.g. related to NSLP-specific objects or NSLP-specific resource management • How should this be structured? • May have additional information in the QSpec • E.g. indicating which parts of the resource request could not be met (such as bandwidth or delay) • See also John’s draft on extensibility • Where should any codes shared between NSLPs be defined?
Other issues • Hop counts • Formatting of objects and message header • Tunnelling case needs further analysis • See draft-shen-nsis-tunnel-00.txt