[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Re: [RRG] Consequences of no renumbering...




>From: brian.e.carpenter@gmail.com

>Two words: fast mapping.

>I missed the bit where it was proved that fast mapping mathematically
>requires aggregatable EIDs.

>As I said a couple of weeks ago:

>However, would a lower estimate, say 2^23
>(8 million) prefixes be an issue? If not, why can't we start on the
>assumption that we know how to support a map with 8 or 10 million
>entries, and have many years to figure out a sparse mapping system
>with several orders of magnitude more entries?

My solution (which I once outlined on this list) does provide such a mapping: The dest. EID is not even evaluated (and btw could therefore be either IPv4 or IPv6 !!!) for doing the forwarding decision.

Instead forwarding is  based on knowing the destination DFZ-router's geographical location ID (square degree geopatch#, square minute geopatch#, square second geopatch#), conveyed somehow (e.g. inside a prepended header, like in LISP). Let's say, a router maintains 3 tables t1 ,t2, t3.

t1 has (90+90) x (180+180) = 64800 elements. t2 and t3 have 60x60 = 3600 elements. Forwarding:

if dest.square degree geopatch# not equal to the one of the current router then

next-hop = t1[dest.square degree geopatch#]

elseif dest.square minute geopatch# not equal to the one of the current router then

next-hop = t1[dest.square minute geopatch#]

elseif dest.square second geopatch# not equal to the one of the current router then

next-hop = t1[dest.square second geopatch#]

else ... /* we are close enough to have a look at the dest. IP address ! Or should we subdivide the square second geopatch once more ? */

Hence it will always be a single table-offset plus eventually 1 or 2 or 3 checks. The dest.IP address needs to be unique just at the destination geopatch. So 4 octets should be sufficient for all times. Billions of billions of users  on earth can hereby even be served.

 

I did  display how to fill these tables. Meanwhile however I have been progressing substantially,  so that e.g. partition-free topologies at whichever zooming level will be computed, or that this is done without the need of configurational data per node (i.e. its highest zooming level to show up), instead, that pre-given i.e. standardized zooming factors will do, and also, unlike PNNI, that any node will well be surrounded first by strict links, then by loose and looser links, i.e. such that e.g. a node close to the rim of a hemisphere won't be immediately surrounded by most loosest links (no uplinks).

 

IMO, this is fastest mapping, a side effect from eliminating the need to build prefixes and/or to do caching

in short from eliminating the scalability problem once and forever

 

Heiner

 

 

 

 

Heiner