[EMAIL PROTECTED] says:
> This reminds me of the comments everyone has made about contortions to
> store mail into AFS when kpop is available and works far better as a
> mail delivery paradigm.
>
> It's not the same thing. It makes little sense to store news in afs,
> because nntp is better at delivering news. But kpop is only suitable for
> delivering mail, not for accessing your old mail that you have stored away.
> Unless you want to use imap (I don't), it makes a lot of sense to store your
> mail in afs.
The mail you have already received should be stored in your home
directory, which presumably should be in AFS space. However, it makes
little sense to make the /usr/spool/mail or /var/spool/mail into an
AFS space thing for a large variety of reasons I've mentioned in
another message. POP works far better for this.
> Once you've decided to store your mail in afs, then it starts to make sense
> to have the delivery agent drop it right into your box rather than having to
> go fetch it. In particular, if you are at the dead end of a slow link, it
> saves having to traverse the link twice.
Well, you should run the POP client (for me, thats MH's inc command)
from a machine on the "main" end of the slow link, not from your home
site. If you do it that way, you'll be fine -- data crosses only
once. In a kerberized environment its easy and safe to bind your
equivalent of inc onto a shell script that invokes a kerberized rsh,
so there is no real problem with it.
> The difference is that news is a huge stream of read-once (if at all) data,
> but mail is a smaller stream that tends to get stored away and re-read at a
> future time.
Partially true. However, I've found that using AFS as a means of
distributing around rapidly changing data of any sort -- be it news,
mail, or anything that alters a lot in the day -- is a bad idea. AFS
works best for large collections of virtually entirely static data, or
for collections of data that's being dealt with almost entirly by one
user. In the former case there is no problem with replication, and in
the latter case there are so many other single points of failure and
the like that it doesn't matter and replication wouldn't help
performance anyway. When you try to use AFS to share rapidly changing
data across an entrprise things tend to break down.
Perry
