The IP over ATM Mailing List Archive by date

[Curtis Villamizar <curtis@ans.net>]

In message <9505111845.AA08270@milliways>, Andrew Smith writes:
> 
> I don't think we should let this thread die until we have answered the
> question posed by Grenville: the WG has never defined what it means by
> what Curtis calls "large". The only concensus we seem to have so far
> is that 2 <= "large" <= less-than-global. Can we please try to get
> a little more focussed on narrowing this range? 
> 
> Andrew


Andrew,

With all respect, I don't think continuing this thread is productive
and I'll say why.  First let me summarize.  (Please correct me if I
misquote anyone).

Grenville suggested that there were two dimensions of "large" when
applied to discussion of "large LIS".  One was the number of hosts.
The other was the geographic size, which then has implications on the
number of switches in the LIS.  

I responded that 1) I was referring to "large" in terms of the number
of hosts in discussion of protocol scaling and 2) that large in terms
of the number of switches and delay between them is a traffic
management issue which I don't think we want to rehash (please stop at
http://engr.ans.net/nap-testing/ if before opening a discussion on
this topic and maybe let do it off the list).

Hiroshi I think first brought up the point that "large" could also
mean the number of LIS in an internetwork.  I pointed out that for
Classical IP over ATM this is not a problem as long as aggregation of
addresses through hierarchical address assignment if being done
effectively and pointed out that an operation area group, CIDRD, is
dealing with this issue on an ongoing basis.

That's the end of the summary.  Now the reason why I don't think we
should go on about this.

The reason I don't think continuing this is highly productive is
because we have identified the dimensions in which LIS and
internetworks do and don't scale well.  So far we have:

	LIS anmd internetwork scaling factors brief summary
  # of hosts in a LIS			doesn't scale well
  geographic dispersion			a traffic management issue
  number of LIS in an internet		so far, so good

  NOTE: So far so good mean we are doing well at 50,000 networks with
  a lousy job of aggregation and X million hosts (4,000,000+ hosts
  estimated at the last DNS survey I looked at), but we need to
  aggregate better to continue to grow.

I don't think it is useful to start arguing over whether the optimal
LIS size is 50, or 100, or 200, or 1,000 hosts.  The point is that as
the LIS grows, some things become problematic, an we've enumerated
some.  If you try to pin a number of this, you get into issues such as
are these 20 Crays or 20 Macintosh cumputers and what mix of protocols
are going to be used.

Curtis

ps- now can we end this thread...