From: Dave Barr (barr@cis.ohio-state.edu)
Date: Fri Jul 10 1998 - 16:54:11 CDT
>On Fri, Jul 10, 1998 at 11:50:50AM -0700, John Stanley wrote:
>Studies show that there are a small number of nodes on USENET through which
>a vast majority of articles reaching almost all sites pass through. If
>these sites happen to be top level signers -- or have a live connection to
>a top level signer (we don't thus have to trust the hub sites) -- then they
>can do collapse as new, non-collapsed articles pass through them.
>
>This involves more collapsing of course, and is not 100% efficient. But 90%
>efficient is adequate.
Looking at Freenet's last report that seems true on the face. You could
probably easily achieve 90% efficient with the 10 top servers.
However the immense added load of looking up every article through
a key database, even if local, and resigning it would I think quickly
bump those top 10 sites down into the 50's or more. (of course the
degree is mere speculation on my part).
At 80 bytes for every art that's still worse than Cancel-Lock,
but admittedly percentage wise merely background noise.
My real worry here is CPU -- with spam filtering and such we're asking
more and more from every server in order to keep up. Add to that
certificate stuff, even if it can be spread out, I fear will really
be the killer.
--Dave