[3339] in Release_7.7_team

home help back first fref pref prev next nref lref last post

Re: Athena Disconnected Operation White Paper Draft 2.

daemon@ATHENA.MIT.EDU (Jonathon Weiss)
Tue Jun 11 12:03:10 2002

Message-Id: <200206110109.VAA08321@localhost>
From: Jonathon Weiss <jweiss@MIT.EDU>
To: lcs@MIT.EDU
cc: Bill Cattey <wdc@MIT.EDU>, source-developers@MIT.EDU, release-team@MIT.EDU
In-reply-to: Your message of "Sat, 25 May 2002 01:26:28 EDT."
             <CMM.0.90.4.1022304388.lcs@defiant.mit.edu> 
Date: Mon, 10 Jun 2002 21:08:13 -0400


Sorry these comments are so late.  I'm trying to catch up on some old
mail.  Ironically, I'm doing this in disconnected mode. :-)

lcs> Lockers are a good model because they define a chunk of (usually)
lcs> network-accessible resource which can be handled by itself.  Instead of
lcs> thinking of which subtrees of /afs to mirror or cache locall, you can
lcs> use the granularity of lockers.

It's worth noting that it't not always worthwhile to bring eerythign
local to the laptop.  For instance, I've brought my inbox and dotfiles
local, but not all of the old mail that I don't need to access
regularly.  In my case this was a compromise against the time and
space needed to bring everythign local.  Certainly there's also no
need to bring the Solaris binaries local onto a linux laptop, just
because they're in the same locker.

lcs> We could modify named to check whether the machine is off the net, at
lcs> which point it responds immediately with the canned "local map".  This
lcs> could implement some useful workarounds, e.g. establish localhost as an
lcs> MX host for "outgoing.mit.edu", so that mailers would still be able to
lcs> send to outgoing to relay their mail but in fact it would be queued
lcs> locally.

Your specific suggestion about making localhost an MX record won't
work, since we don't run a sendmail listenter by default (one less
network service to have holes).  This doesn't matter for most
applications, because they generally invoke sendmail directly to queue
the mail.  I guess netscape is the known exception, and I don't recall
how evolution behaves.  If all else fails I suppose we could look into
starting an SMTP listener only on the localhost interface, but I don't
know how easy that would be to set up.

dryfoo> We have one advantage with AFS, I think: that when the user is
dryfoo> disconnected, AFS has "lock" among its permission types.  So if the user
dryfoo> is off-line somewhere modifying his own files, his own locker at least
dryfoo> could be locked to all other users who might have modification rights to
dryfoo> any subdirs in it.

'lock' doesn't mean what you think it does here.  It is intended for
temporary locking of files to prevent multiple clients from changing
the same file simultaneously.  I'm pretty sure that a client's lock
will be lost if the client falls off the net.

	Jonathon


home help back first fref pref prev next nref lref last post