> I would consider using Terminal Services to deliver the applications
> to the remote sites. It requires no change to your code and performs
> acceptably even on a dialup connection. Of course it requires a
> working connection (but so does a client-server model). It also opens
> up the door for cross platform as official rdp clients are available
> for windows and mac and unofficial clients are available for linux.
> There are some licensing costs for this model.

I have to agree here. A client of mine has several branches all
connecting over a relatively slow wide area network to central
terminal servers. We've been doing it for years and it works
very well. I provide remote support via terminal services session
over a SSH tunnel, one or two other users come in from the 'net via
a VPN tunnel.

If a link drops out the terminal server session remains running,
so all the user does is reconnect to their previous session and
keep working from where they were before. No lost data, minimal
lost time.


Incidently, the client recently had someone demonstrate the use
of a Linux server as a terminal server - not the conventional
X-windows arrangement, but actually as a server that Windows
RDP clients could connect to. They setup Wine to run my Windows
based database application and used OpenOffice, Firefox and
Thunderbird directly on Linux.

It worked... mostly. Apparently some of the thin client hardware
had some difficulties. Also the fonts available in Wine did not
exactly match what I had been using in my database application so
the form layouts did not match exactly. Most of this is probably
able to fixed or worked around - but the client decided to put
off doing anything for a while at least.


--
Geoff Worboys
Telesis Computing