subIP ( PPVPN,TE,IPO ) draft-hummel-ppvpn-tunnel-systems-00.txt

[Hummel Heinrich <Heinrich.Hummel@icn.siemens.de>]

NAME of I-D:

http://search.ietf.org/internet-drafts/draft-hummel-ppvpn-tunnel-systems-00.txt
SUMMARY:
This draft shows how n sites of a private network may be
   interconnected by n-1 shortest (L2/L3)-tunnels that form a minimal
   tree tunnel system, or by n fairly shortest tunnels that form a ring
   tunnel system. Algorithms are provided for computing such tree/ring
   tunnel systems within a fraction of a second. Analogous variants
   thereof may be used to determine tree/ring tunnel systems for
   interconnecting all member VPNs of a VPN alliance.  Tunnel weights
   may either pursue the optimal use of carrier resources or the minimal
   monetary costs for the customer.  Explicit routing mechanisms are
   proposed for establishing/ testing/ tearing down entire VPN-tunnel
   systems of arbitrary topology (trees, rings, meshes).
 

RELATED DOCUMENTS:

http://search.ietf.org/internet-drafts/draft-tsenevir-l2vpn-pmesh-00.txt
 

WHERE DOES IT FIT IN THE PICTURE OF THE SUB-IP WORK

The ID presented fits into

- PPVPN WG (the application is VPN)
- TE       (the application may be Closed User Group (CUG)or
            Virtual Service Provider Network (VSPN) )
- ipo      ( optical rings ! ) 

WHY IS IT TARGETED AT THIS WG(s)
See added text in brackets.
Additonally w.r.t. ppvpn:
Since I realized that all existing VPN models do not exclude "partial mesh" such
optimized/minimal tunnel systems may also be integrated in these models (though their descriptions
are full with tunnel sharing between different VPNs rather than with utilizing the own tunnels also
for transitting some sites) 
 


JUSTIFICATION

The number of inter-site tunnels can be reduced from O(N-square) to O(N).
E.g. a tunnel ring between N sites,needs N tunnels and not N*(N-1) tunnels (full mesh).
If N=101, than 9999 tunnels can be saved which is 99 %.

The pretty good tree/ring computation algorithms as outlined in the draft (sec.2.1, 2.2) take 
less than 1 second even if N=100.
For comparison:The absolut best ring computation (= notorious travelling salesman problem)
would take 10 to the power of 86 years, if the computer is able to check each arrangement
of 100 sites within 1 micro second.