STARLAN NAU connections on UNIX pc, some comments ...

[Lenny Tropiano]

As most of the readers of these groups know, I just got myself two STARLAN
NAU boards.   I was having some problems with configuration of the HDB
files (Devices, Dialers, etc..) because my installation script didn't
adjust to my own files here.   Thanks to a few people who replied to my
"plea" for help (since I had no manuals -- and didn't want to spend hours
trying figure out something that was a matter of changing one line...).

Well after some more investigation, I noticed why my uucico didn't want
to dial out using the starlan device.  What happened I had a special
version of HDB here that didn't have the starlan calls (starlan library)
compiled with it.  It was trying to open up /dev/starlan and just
send stuff to it, expecting login: in return.  That obviously didn't work.

Some comments about STARLAN connections, they are real fast.  It's nice
to be able to "cu" to my machine (which previously just had one 19.2K 
RS232 connection between it).  And then do uucp transfers at the same
time (in either direction) as well as "cu" again, or "cu" the other
way, all simultaneously ...  The degradation in speed is negligible,
but the load on both machines in my small network were pretty minimal.
Thad Floryan (and others) will be able to tell us how things work when 
there is more than one STARLAN node connected in daisy-chain fashion.
I'd like to hear if anyone actually has a *star* configuration using
the STARLAN NEU (Network Extension Unit) [the Universities might
do it this way...]

Establishing a network connection using "cu" does take some time
to sync up and actually connect to the remote site.  I wonder why
this is the case, uucico's start up real fast.   I think because the
"sl" (starlan login) service has to handle "pty" type devices called
"psx" it might take a few seconds (almost 30) to get connected.
I don't know how the Asynchronous Emulator (from the Phone Manager) works
because I keep getting this:

                 +--------------------------------------------+
                 | Error                                      |
                 |   Profiles                                 |
                 |     /u/lenny/Filecabinet/Profiles/-1 not   |
                 |     found in Filecabinet/Profiles          |
                 |                                            |
                 +--------------------------------------------+

My device is set up as "STARLAN", with the data profile as STARLAN.

-rw-rw----  1 lenny   icus        844 Oct 11 18:37 Profiles/STARLAN:Sl

Well so much for that, I'm sure I'll figure it out, or at least have 
someone tell me what to do... ;-)

Ok, a little benchmarking ...  The uucp protocol by default for STARLAN
connections is the "e-protocol" (error free data path).  I would assume
there is no, or very little, handshaking done here.  The data path is
assumed to be error free, and 1024 packets are shoveled down the line to
the remote node.

I needed a *big* file to test this with...  A fairly large one was my
news history file ... 

-rw-r--r--  1 news    news    2352055 Oct 14 23:27 /usr/lib/news/history

Using the "e-protocol" this file went to my other machine, icusdvlp over
STARLAN this fast!  

icusdvlp!root M (10/15-0:24:58) (C,2252,1) [starlan] -> 2352055 / 78.550 secs,
29943 bytes/sec

Yes, 78 seconds over 2MB of file...  Yes, and at 29943 bytes per second!
That's bytes per second!  I get about 1100 bytes per second on the average
for a 19.2K Telebit Trailblazer connection!

Then I switched to the "g-protocol" using a WINDOWS=7 uucico, and still
using STARLAN transport medium.

icusdvlp!root M (10/15-0:43:48) (C,2268,1) [starlan] -> 2352055 / 1045.800 secs,
2249 bytes/sec

This took, about 17 minutes, quite a bit of a change.  Now you see how
much overhead the g-protocol uses.  But still here it's twice as fast
as my Telebit (and that's fast), and still faster than the 19.2K direct
RS-232 connection, which was ~32 minutes (at WINDOWS=7)

icusdvlp!root M (10/15-1:14:48) (C,2273,1) [tty000] -> 2352055 / 1942.834 secs,
1211 bytes/sec

So I guess it's probably good to say that 78 seconds vs. 32 minutes is quite
an improvement!

Now what's in store for the future of my STARLAN.  Well I'd like to write 
my own server.  Maybe implement TCP/IP over the STARLAN ... I'd like to
write a better "cu" for the thing, it's too slow to connect.  Sending
character by character isn't good either.  Adding new services seems
simple (even though I don't have a manual).  Each service looks like it
has an unique number, located in the /usr/net/nls/starlan/dbf file:

1:na:NULL,:/usr/net/servers/sl  # Remote login
102:n:NULL,:/usr/net/servers/uucplogin  # Mail/File transfer

(This case it's 1 and 102).  I could easily add another program, to
do TCP/IP connections... 

Well I babbled on for long enough ... Take it easy all.

							-Lenny
-- 
| Lenny Tropiano            ICUS Software Systems      [w] +1 (516) 589-7930 |
| lenny at icus.islp.ny.us     Telex; 154232428 ICUS      [h] +1 (516) 968-8576 |
| {ames,pacbell,decuac,hombre,sbcs,attctc}!icus!lenny     attmail!icus!lenny |
+------- ICUS Software Systems -- PO Box 1;  Islip Terrace, NY  11752 -------+