Re: [9fans] Observation about sharing network/grid

[Jim Choate <ravage@einstein.ssz.com>]

On Sun, 5 Oct 2003, Charles Forsyth wrote:

> still, you're right: NFS could have done that, and perhaps the suenix
> re-implementations do.

NFS has been doing that for at least two yeas on Solaris, HP/UX, and AIX.
Might want to review the man pages. When I used to work for Tivoli we had
to make sure this was turned off because of some low level sys call
conflicts with Tivoli during installs from NFS images. Wasn't a big deal
since most sites didn't see it as very usefull.

In a public namespace/filesystem context however the importance changes
dramatically. If you don't allow transitivity in the mounts (and
lazy-update is another must have in the real world) then the source
servers get completely bogged down. Just imagine trying to export the
Gutenburg Project by forcing everyone to go to the project itself. By
using this technology a handfull of sites around the country could make
the primary connections and then export those out to their immediate
region. Then as users hit the second level server, only files that have
never been hit before (or have changed content even one bit) need be
re-propogated.

This whole 'a few sites share to another set of few sites' is the heart of
'small world networks'. It allows the network to scale without traffic
growing with it, provided the number of connections per node is not too
small or too large. Sort of like Goldilocks ;)


 -- --

God exists because mathematics is consistent, and the Devil exist because we
can't prove it.
                          Andre Weil, in H. Eves, Mathematical Circles Adieu

      ravage@ssz.com                            jchoate@open-forge.com
      www.ssz.com                               www.open-forge.com