Re: [Gnu-arch-users] Ruminations on Arch Desiderata

[Paul Snively]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi, Florian!

On Tuesday, September 16, 2003, at 01:10  PM, Florian Weimer wrote:

> There was some discussion on this previously.  I did some research,
> and it appears that the service discovery mechanism currently
> standardized by the IETF do not fulfill the needs of GNU arch, due to
> its multi-protocol support.  Maybe DNS-SD/Rendezvous is better in this
> regard, I haven't looked at it.
No, this is definitely an issue: presumably arch itself has no way of 
knowing whether an archive you create is publicly accessible or not due 
precisely to not caring whether it's on an FTP site, a web site, etc. 
So "public registry via DNS-SD" is clearly an orthogonal issue. In 
addition to that, "services" are presumed to own their port, but it 
clearly doesn't make sense to say "there's an arch service on port 
XXX," when the port in question is an HTTP port, FTP port, SSH port...

Conversely, on the receiving side, we might consider a daemon process 
that listens for new repositories and, when it finds them, "registers" 
them (in the arch sense) automagically, vs. requiring the user to run 
tla to discover them. OTOH, perhaps it would suffice to have "tla 
archives" lazily search the space, although that seems to have a lot of 
potential for performance issues.

The point is that this is exactly the kind of conversation I want to 
have: here's an idea. What are the issues surrounding implementing the 
idea? Are they surmountable? If not, why not, and what might we do 
instead?

> Anyway, I think it's premature to standardize a naming/directory
> service for arch archives at the moment.  At least for me, the most
> relevant usage patterns are not yet clear enough.  Some more
> experience in this area should enable us to come up with better
> requirements.
Experience in what area? Using arch, using naming/directory 
services...? It seems to me fairly apparent that, while having a 
"global distributed namespace" is nice, having to define names in the 
namespace by hand doesn't scale. Given that there's a technology that's 
included in one popular OS, is available for others, and has been 
deployed in at least one CVS client/server pair, it seems worthy to at 
least consider how arch might follow suit. Note that I'm not saying 
"accept me as a commiter or else;" I'm merely saying that a reasonable 
discussion as to how such an endeavor might (or might not!) work would, 
I think, be beneficial to the arch community. At least, it would 
certainly be appreciated by me.

Best regards,
Paul

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (Darwin)

iEYEARECAAYFAj9n19UACgkQugPBK9DeterX9QCfZ1YGFEfdkQQmdDeKbhF6ySFU
0M4AoKrAquehNttFVp9wa2HzdU/SKOtn
=DAcv
-----END PGP SIGNATURE-----