70 likes | 206 Views
NSIS Flow ID and packet classification issues. <draft-cheng-nsis-flowid-issues-02.txt>. Hong Cheng, Qijie Huang, Takako Sanda, Toyoki Ue IETF#64 Nov, 2005. Motivation & goal. Address issues of packet classification information signaling MRI is not adequate
E N D
NSIS Flow ID andpacket classification issues <draft-cheng-nsis-flowid-issues-02.txt> Hong Cheng, Qijie Huang, Takako Sanda, Toyoki Ue IETF#64 Nov, 2005
Motivation & goal • Address issues of packet classification information signaling • MRI is not adequate • To achieve a general solution for different NSLPs
MRI & Packet Classification • A separate packet classifier is needed above NTLP • The MRI does not always contain adequate information for both routing and packet classification • Packet classification information covers winder than MRI • E.g. Address wildcarding; a range of ports, etc • Packet classification needs more specific information than MRI provides • E.g. It needs to use upper layer info which is not in MRI • NSLP Application may want to signal some flow not indicated by MRI • E.g. Address than cannot be wildcarded, but they will pass the same NSLP node; aggregation; or off-path case
Why PACKET_CLASSIFIER object is not enough • Issues for the current PACKET_CLASSIFIER in QoS NSLP • It is too coarse • only simply specify which field of the MRI to use • e.g. cannot indicate a range of port address, or addresses wildcarding • It does not provide any extra information • everything is still derived from from MRI
Extended classifier object • more specific information is needed • an extended classifier object is needed • It is generally useful for all NSLPs • The signaling of more specific packet classification information is useful for other application • It should be made generic (not NSLP application specific) • The extended classifier object does not need always be present • for efficiency considerations (MRI is adequate sometimes) • an extended_classifier flag could be used to indicate its presence
extended classifier object issues • NAT traversal issues for the classifier object • Difference NAT requirements when placed at difference layer • NSLP is the suitable place • tradeoff between NAT complexity and signaling overhead, reusability • Consistent rule has to be applied • same rules to apply on MRI and the classifier object
Next step and open issues • An extra/extended classifier object is needed (besides MRI) • How should the issue be addressed? • A general NSLP solution? • To be solved separately in different NSLP applications • More than one classifier object? • Signal more than one address pair (not from MRI) • Stacking of classifiers • when QSPEC stacking is required (and classifier is not in QSPEC) • Does this needs to be supported?