Re: [DNA] A Merging Proposal

[James Kempf <kempf@docomolabs-usa.com>]

> Right now, DNAv6 doesn't manage the Complete Prefix List (CPL).
> While it manages the Prefix List with DNAHostPrefixList, it doesn't
> concern whether the Prefix List is complete or not.
>
> We can combine CPL & DNAv6 to make a host always manage
> not only the Prefix List but also its COMPLETENESS. So roughly
> DNA host checks for link change as below.
>
> First, a host manages the CPL for an each link.
>
> Second, when the host moves to a new link,
>
> Case 1. When the host WITH the CPL
> - The host compares the CPL with the Prefix List in an incoming RA
> and assumes a link change if and only if
> there is no common prefix between two.
> (Take notice that it doesn't matter whether the host moves to
> a DNA link or a non-DNA link.)
>
> Case 2. When the host WITHOUT a CPL moves to a DNA link,
> - The host should send an RS with a Landmark prefix.
> - Upon receiving the RS,
> - If a DNA router can send a CompleteRA,
> it sends the CompleteRA
> to provide the (possibly new) CPL.
> Upon receiving the RA, the host can check for link change
> by comparing its Prefix List with the CPL in the RA.
> (Even when a DNA router can send a CPL,
>  it may simply reply with the Landmark prefix
>  in case of no link change to save bandwidth.)
> - If the router can't sends a CompleteRA
> (because of too many prefixes),
> it replies with the Landmark Option with YES/ NO indication.
> The host check for link change with the Landmark Option.
>

Why would a DNA host not have CPL? How would it know if it had CPL or not?

        jak

> Case 3. When the host WITHOUT a CPL moves to a non-DNA link,
> - If the host happens to move to a non-DNA link without a CPL,
> it checks for link change as of CPL
> by performing (multiple) RS/ RA exchanges to generate the (new) CPL
>
> Also LinkID prefix makes that any two RAs in a link have
> at least one common prefix to assure graceful prefix addition or removal.
> There would be no false decision, even when there is a temporary
> asynchronization among DNARouterPrefixList.
>
> In this way, I think we can integrate CPL and DNAv6 in an efficient way.
>
> I elaborated the above further as below.
>
> 1. DNA Host Operation
>
> 1.1 CPL management.
> 1) A host generates the Prefix List of a currently attached link.
>
> 2) The host declares the existing Prefix List is complete when
> - it receives a CompleteRA
> (as of DNAv6 in a DNA link where a router can send a CompleteRA.)
> or
> - it performs RS/ RA exchange num_RS_RA times
> (as of CPL either in a non-DNA link or in a DNA link where a router
> can't send the CompleteRA)
> - The Prefix List needs a complete flag to indicate its completeness.
> Currently, in CPL, Candidate Link Object uses the number of
> RS/ RA exchanges to indicate its completeness.
>
> 1.2 Link Identification.
> 1) Assume a host receives a hint of a possible link change.
>
> 2) Then the host sends an RS with a Landmark Option.
>  * If the host has the CPL, I think it can omit the Landmark
>  without degrading performance.
>
> 3) Upon receiving an RA with at least one prefix,
>  the host checks for link change as below.
>
> IF the host has the CPL (Complete Prefix List)
>    the host assumes a link change
>    if and only if
>    its Prefix List and the Prefix List from the incoming RA
>    have no common prefix.
>
> ELSE {
>          IF it receives the CompleteRA,
>               the host assumes a link change
>               if and only if
>               its Prefix List and the Prefix List from the incoming RA
>               have no common prefix.
>
>          ELSE {
>                     IF it receives the RA with the Landmark Option
> which it sent
>                          the host assumes a link change
>                          if and only if
>                          the Landmark Option indicates a link change with 
> NO
>
>                     ELSE
>                             the host checks for link change as of CPL.
>                             (i.e. it performs RS/ RA exchange num_RS_RA 
> times
>                             to generates a (new) Prefix List and
>                             compare it with the existing one.)
>          }
> }
>
> * I guess we can make Sec 5.2.5 of DNAv6 more clear with the above.
>
> 2. DNA Router Operation
>
> 2.1 CPL Management
> 1) A router generates the CPL (Complete Prefix List) by performing
>  multiple RS/ RA exchanges as of DNAv6.
>
> 2) The router includes the LinkID Prefix as of DNAv6
>  to assure that two RAs always have a common prefix.
>  * With this, we can add or remove a prefix without resulting in
>  false link identification.
>
> 2.2 Link Identification.
> 1) Upon receiving an RS, the router sends back the CompleteRA,
>  if possible.
>
> 2) If not possible, the router indicate a link identity
>  with Landmark Option with YES/ NO bit.
>  * I think we may do away with YES bit because we simply put
>  the Landmark prefix in PIO to indicate YES.
>
> 3) Or we may make a modification to save bandwidth as below.
>
> The router first checks the Landmark Option in an RS and
>
> IF the Landmark is in the same link,
> the router simply sends back the Landmark Option with YES.
>
> ELSE {
>          IF possible, the router sends the CompleteRA.
>              * because the host needs the CPL anyway.
>          ELSE the router sends back the Landmark Option with NO.
> }
>