dovich@ge-dab.UUCP (Steven J. Dovich) (07/08/87)
We seem to have run into a large difference of opinion on how to deal
with net growth. The whole problem boils down to maximizing the flow
through a graph, where each node and edge has a capacity value. The
values for each node and edge are determined by the owners and
administrators of the site(s).
Mr. Webber has taken the time to present his case for establishing
transport volume quotas on links. Bob's approach will work, no
argument here. My concerns are what it means to the distribution
mechanisms.
The quota solution increases flow by discarding traffic at congestion
points, letting uncongested portions of the graph add traffic until
the region is at capacity. The topology of the net is such that the
congestion points are those which interconnect sizable subgraphs. The
net effect is to lose deterministic expectation of distribution. You
want justification for this? Mr. Webber thinks much of the traffic
could be better answered by utilizing other resources.
By analogy, suppose my daily newspaper grew to about 10 lbs an issue.
My carrier might get feel that the load was too much, especially
since there are several neighbors who also get a copy. The solution
is not to discard some pages until the load is light enough to handle.
I might want to see the articles which were discarded. The better
solution is to get another carrier to deliver part of the news, a
fixed (probably topical) portion of the news. That way none of my
carriers has to deal with the total load that I might be interested
in.
The key point here has been mentioned before, some sites have
resource contention for communications traffic, and other sites have
news storage problems. There are many different problems which limit
the ability to pass news. The solution proposed by the backbone (they
will be disappointed if I don't refer to them as a Cabal) is to
establish several alternative backbones which carry whatever traffic
their resources (and management) can bear. Those sites which can
afford more news traffic can recieve news from several distribution
backbones. Those whose communication resources are limited may not
be able to pass multiple distributions.
This also is a workable solution since it also has the effect of
increasing the flow of the network graph. The total traffic is
separated and classified with a distribution tag. Each node of the
graph should pass only complete distributions (a simplification for
analysis), hence flow is optimized by developing additional spanning
trees which carry well-defined distribution classes. Each of the
spanning trees should have enough internal redundancy to ensure that
news will propogate despite failure. With current network
connections, there is sufficient connectivity to provide another
distribution without requiring additional connections.
The difference between the proposals is this, the quota plan presumes
that we have already reached the limits of resource availability, and
need to establish limitations at existing congestion points. The
alternative backbone plan expects additional resources becoming
available to support traffic. Again growth plans must also account
for changes in distribution characteristics. If the net doesn't leap
at nondeterministic distribution, the quota plan will not succeed. My
vote is cast for determinism.
The "alt" and "inet" distributiona groups are developing to provide
alternative backbones. Admittedly, they are not spanning trees of the
entire net. But sites interested in carrying those groups can find a
feed if they are willing to bear the costs involved. Each of the
standard categories should also be considered a distribution. The
current backbone just happens to carry 7 distributions (comp, misc,
news, rec, sci, soc, and talk).
If more information were available about each of the distributions,
then each site could make their own determination about where to spend
their news resources. Perhaps the other backbone Cabals (alt and inet)
can publish periodic info. I consider info on "alternate" newsgroups,
and distribution backbones particularly appropriate.
--
Steven J. Dovich <dovich@ge-dab.GE.COM>
General Electric Company
UUCP: ...!mcnc!codas!ge-dab!dovich 1800 Volusia Ave, Rm 4333
Voice: +1 904 239 2564 Daytona Beach, FL 32015