I have left off a clause that makes my point of view inconsistent. My comments regarding geographic correlation were with regards to a specific [example of a] provider's network allocation. On a provider by provider basis, topology does, and will increasingly, match geography. Networks that span large geographic areas should be deaggregated into regional ares such that aggregation can be efficiently implemented. Additionally, within this provider, they should allocate their chunk of the address space in a geographic manner. So you have correlation at the high level (providers in an area) and at a low level (particular provider's IP addressing in a small area). I did not mean to [though I certainly did] imply that 2 separate providers would have a strong correlation of IP address allocation in a locallized geographic region. Let's take these examples: KINDISP has a network in 4 continents. They obtain 4 separate sets of network space from the corresponding NICs. They then develop an allocation plan [ignoring crystal balls, but utilizing inferences from business plans and historical trends] and allocate netblocks in a regional manner. DUMBISP had a network in 1 continent. They expand into 3 more and continue to use their same original netblock. They allocate addresses chronologically, instead of geographically. This continental allocation is being done today, the NSP's internal allocation is what I was attempting to focus on. -alan Quoting Paul Ferguson (ferguson@cisco.com):
At 04:35 PM 11/6/97 -0500, Alan Hannan wrote:
The lack of correlation is the exception, in my experience, than the rule.
I'm not sure I agree, but that's not really the point.
I believe the point is that we should not assume that they are one and the same. Topology (or as they say in IPv6-speak, aggregator, next-level-aggregator, etc.) is quite critical in maintaining sufficient levels of aggregation.
- paul