Hi Gavin -
Yep..you are right on all counts. In a nutshell, the way it works is that
when jinetd determines that a Telnet connection is to be established, it
calls a procedure, PTD_INIT (PTD stands for Pseudo Terminal Device...kinda like
pty, but not ..;-)), that does all sorts of stuff, including allocating the
device, grabbing system buffer resources and initializing the Telnet connection
with the TCP layer. Right before PTD_INIT exits, it calls a procedure that
notifies JOB/SESSIOIN management that a session needs to be started on this
newly created device. After this, PTD_INIT exits and INETD is now out of the
picture.
The feedback that the Telnet team got when doing the Telnet/iX server
design was that an extra server process, as with VT, was a BAD idea, so we
implemented the driver in a way so that it is directly callable from the
CI. The way to view the Telnet driver is that it is a callable library that
the CI and JSMAIN have enough intelligence about in order to get and receive
data properly. So, if you would look at a process tree for a Telnet connection,
all you would see at the bottom was the CI process.
Whew...that's my 30 second Telnet tutorial for today...have a nice Labor Day
weekend for all of you in the US and for our no-US compatriots, have a nice
weekend also.
Cheers from the blistering midwest and Lawrence, Kansas, the hottest place
in the lower 48...
Jeff B.
Gavin Scott ([log in to unmask]) wrote:
: Dave writes:
: > I've always assumed VT-MGR to be faster because it is an OS type process
: > where TELNET or FTP is running in a JOB which is dependent on the
: > job queue settings and such.
: This is only somewhat valid for the connection establishment part of the
: process. Even though the telnet "listener" is part of INETD, once you start
: to establish a telnet connection "magic" happens and telnet is just as much
: of an "OS type process" as VT from that point on.
: You can, for example, abort the INETD job and while this will prevent new
: telnet sessions from connecting, it won't affect any which are already
: connected.
: Telnet also seems to manage not to need a separate "server" process for each
: session (unlike VT which has a VTSERVER process for each connection). I
: *think* this means that telnet connections require one less process per user
: than VT sessions.
: G.
--
----------------------------------------------------------
Jeff Bandle
Hewlett-Packard email: [log in to unmask]
Commercial Systems Division Voice: (785) 842-0369
Remote office: Fax: (785) 842-0273
4701 Harvard Dr.
Lawrence, KS 66049
Disclaimer: Comments are my own, not HP's
----------------------------------------------------------
|