| [email protected] writes:
[...]
| > I see a problem coming here: how does an unreplicated document
(say my
| > own home page) mane a reference to such a replicated document?
If I
| > have a reference to the closest replica, a user far away who
follows
| > such a link will get pointed to the replica closest to *me*, not
| > closest to her.
| >
| > Some possible solutions:
| >
| > - a translation scheme whereby clients "know" (e.g. from a local
| > configuration file that may be updated automatically as mirror
sites
| > are added) that information at host X is identical to info at
host Y
Yes, this is possible with the 2.0 library. The line mode client
doesn't have the command line option yet, but the client library
can use a rule file, just like a server.
| > - a magic string in hostnames that is translated dependent on the
| > geographical position of the client (e.g.
| > http://info-cern.closestmirror/...)
This has been discussed, in fact having a host name which translates
info many IP addresses -- all the apparatus is there already in DNS
and most people say it won't break anything, we just need code to, if
DNS returns >1 IP address, ping them all to get the closest and
remember which one it is.
| > - upon first contact with a server, it might respond with "please
| > try the following mirror site which is closer to you" (this could
be
| > put in HTTP2 I suppose).
This is already the HTTP2 spec. The reply field can be a "forward"
reply containing a pointer to the real object. That is, HTTP2
servers can be used as name servers. No code yet in the library.
See http://info.cern.ch/hypertext/WWW/Protocols/HTTP/HTRESP.html#z9
| > This is a real problem with embedding location information in
URLs...
|
| Sounds like it's time to move URN's (or whatever persistent
Internet
| resource identifiers are being called these days) out of the theory
| stage and into practice.... anyone know what the status of the URN
| work that was/is apparently going somewhere in the IETF?
I thought it was going to get closer, but basically it was agreed
that the format urn:publisher/id would work with a finite number of
publishers. I think you might see Peter Deutch and Chis Weider maybe
putting some code together???
Remember that if you have to contact a name server every time,
you slow things down anyway. Basically, a little common sense
in the client would guess that a related document was avalable from
the same server as last time.
Tim