[comp.protocols.tcp-ip] Making terminal servers just like internal serial ports

garvey@johnny5.uucp (Joe Garvey) (03/15/91)

It seems to me that with all the Sun accessories that live off the SCSI
bus, that someone in the terminal server market should be doing something
similar.

I'd have loved to have a terminal server, where I had actual device files
on my system to address individual serial ports.

It seems to me that you could do this with psuedo-ttys and a daemon that
monitored the slave side of these devices.

The real problem would come when you needed so set the baud rate or
handshaking mode. The actual flow of bits would be pretty simple.

Why doesn't anyone do this? Is there a protocol under developement?

I have seen an add for the Annex III. It claims something like this...
anyone know more?


-- 

Joe Garvey                       uucp: sumax!quick!johnny5!garvey
J5 Research                      map entries are wrong for johnny5. They're
Bothell, Wa.                     being fixed. Please use address above.

griff@Xylogics.COM (Scott Griffiths) (03/15/91)

I tried to respond to Joe Garvey directly but the mail kept bouncing so I
decided to post this here.

For some time now, the Xylogics family of terminal servers has provided a host
based utility called rtelnet (reverse telnet). This program allows the user 
to specify a /dev/device_name entry for use with applications which are used 
to talking to /dev/tty types of devices. A data path is set up between 
the host and a specified port on the terminal server so these applications 
work without modification. The most often used applications are for printers 
and modems on the network (lpd, cu, tip and uucp).

The utility actually works as follows. A free pty is found and reserved for
use (by opening the master side). The user specified device name is linked to 
the corresponding slave device. Finally, a telnet connection is opened between
the host and the specified terminal server port.

Please let me know if you need any more information on this capability.

Scott Griffiths					phone: (617) 272-8140
Xylogics Annex Technical Support		email: griff@xylogics.com
-- 

Scott Griffiths					phone: (617) 272-8140
Xylogics Annex Technical Support		email: griff@xylogics.com

john@loverso.leom.ma.us (John Robert LoVerso) (03/16/91)

> Why doesn't anyone do this? Is there a protocol under developement?

Some terminal server vendors do let you do this.  There are several ways.
The easiest involves something on the order of a user-level "reverse"
telnet daemon that runs on your host, connects to a pty pair, and makes
a telnet connection to the terminal server.  This gives you a device
that you can poke random programs at.  Xylogics provides source to such
a program with the software distribution that comes with the Annex.
cisco provides a similiar program via anonymous ftp access.  One or
two other such programs have been posted to comp.sources in the past.

The pty approach has several hazzards, as it stretches the abilities
of the pty mechanism in various ways.  However, it is a good 90% solution.
Of course, it can reek havoc with the performance of your machine - as
can anything based upon the traditional BSD telnetd/pty interaction.
Using the "in-kernel" telnetd support code can remove this performance
problem.

At least one company (Artecon) resells [Annex] terminal serves with a
SunOS device driver that does the telnet connection in the kernel and
links directly to the tty stack.  This supposedly removes several of the
problems with the pty approach, but I've never used it, so I can't comment
further.

John