Flow Label Discussion

1
Flow Label Discussion

• there have been a number of proposals to pin down
  the specification of the Flow Label field
• choices
• complete standardization of the original spec
  (currently in appendix of RFC2460)
• specify and standardize a different usage
• continue to leave it unspecified
• should decide soon, even if decision is (3)

2
Todays Discussion

• due to poor planning, only three proposals will
  be presented today
• Deering RFC 2460, Appendix A approach
• Conta providing protocolport info via Flow
  Label
• Bound hybrid approach
• 10 minutes each
• expect discussion to continue on list and at
  interim meeting

3
RFC 2460, Appendix A

• Problem Statement
• intserv style QoS requires routers to identify
  flows (either individual or aggregates) at a
  granularity finer than source/destination addr
  pair
• (note flows are not the same as difserv classes)
• the technique used by IPv4 (looking at ports) is
  potentially much more expensive in IPv6, because
  of header chaining
• besides, routers looking at ports is a layer
  violation, and doesnt work if ESP encrypted

4
RFC 2460, Appendix A (cont.)

• Why this solution?
• alternative of hop-by-hop tags / labels (like ATM
  or MPLS) has significant problems with multicast
  flows over multiaccess links, like Ethernet.
• using source address flow label as flow key
  avoids need for any flow allocation protocol
• using random flow labels enables more efficient
  lookups than otherwise (but not as easy as
  hop-by-hop labels)