.TL
Installing RJE on Unix: A Case History
.AF "Intermetrics"
.AU "Jude Miller"
.MT 4
.HU "Introduction"
One of our main goals in acquiring Unix was to
setup a convenient telephone link to IBM MVS
via JES2, 
a descendent of HASP.  We allowed a month for this goal,
but required closer to five months to achieve it.
This history is presented with the hope that someone
may save time by not repeating all of our mistakes.
.HU "Configuration"
Our first mistake was in ordering our hardware.
We ordered a PDP-11/70 with 128K words of memory,
two RP06 disks, DH11 multiplexer, terminals, printer,
and mag tape.  We also ordered a DQS-11B, a Bell 208
modem, and an additional phone line.  However, the DQS-11
synchronous line controller was quoted as an 8 month
delivery, so a DQ-11 was ordered instead.  Unfortunately,
a DQ-11 is not a DQS-11.  Although they are designed to do the same job,
they do it in different ways.  It may be possible to
construct a DQ-11 device handler which makes it possible to
simulate  a DQS-11, but we were unable to do so.
So, we ordered a DQS-11 after all.  Fortunately, it was shipped within 45
days.
.HU "Hardware Installation"
Our DQS11-BA (Bx -- EBCDIC, xA -- EIA signals) was
quickly installed and checked out by DEC personnel.
Unfortunately, functional test of the DQS-11 requires two
DQS-11's.  Because the DQS-11 is a DEC Special Systems Group product,
the local DEC office was not equipped to completely verify its operation.
A diagnostic (DEC-supplied) indicated that the device was okay
internally, but did not check the actual modem signals.  We decided to
assume these were okay, and proceeded to install Unix 'hasp'
according to "I.1  Setting Up PWB/Unix."  
.P
After building the hasp system via the supplied 'make' file,
We tried the command 'haspinit'.  The message 'phone connect not made'
was printed.  After a few days of experiments and code reading, we
determined that the file 'rje/lines' was not correctly initialized.
In our case, it needed to end with our remote number followed by any
non-delimiting character followed by plus.  The plus indicates that no
autodialing should be attempted.
.P
We confidently modified our 'lines' file and retried haspinit.
Many 'stray interrupt' messages began appearing on the system tty.
DEC was called , and a different CE opined that the device address for
the DQS-11 had been inappropriately chosen.  The address was modified
to 765050, the system description file was updated, and the system remade.
Fresh trials of haspinit produced only the message 'phone connect not made.'
After a few hours of this, we inserted a patch cable between the modem
and the DQS-11 and monitored traffic with a logic analyzer.  'ENQ' messages
were being transmitted to the modem, but no 'ACK's or anything else were
being received.  Eventually, someone noted that the modem status lights
indicated transmission when the patch cable was connected, but not when
the normal cable was.  We followed this clue and discovered that that the
cable terminated at the DQS-11 end at a circuit board holding several
wire jumpers, none of them cut.  We had the prints handy for the DQS-11,
and found comments for installing this cable on a DQ-11, but none for a
DQS-11.  The DQ-11 instructions required several jumpers to be cut, which
we severed with some glee.
.HU "Software Installation"
At this point, we were able to run haspinit to completion.
We immediately used send to transmit a few jobs to MVS and waited
eagerly for the results.  When it became apparent that none were
forthcoming, we signed onto MVS through another RJE system and found
the results of our transmissions: a few incomplete JOB cards and JCL
error messages.  We consulted another installation with functioning RJE.
They said that they had had trouble with their MVS host's remote line
definitions.  Our remote unit was indeed incorrectly defined. The correct
definitions included:
.DS 1


RMTxx 2770 ABUFEX CONSOLE DISCINTV=0 FIXED LINE=yy
	NUMPR=1 NUMPU=1 NUMRD=1 ROUTE TRANSP BLOCKED
LINEyy CODEA HDUPLEX HISPED IFACEA NOADIS TRANSP EBCDIC


.DE
.HU "Nits"
The long-awaited time was at hand; jobs began to flow from Unix
to MVS and back again.  Our users soon discovered two problems, though.
The first was that our MVS host used non-standard EBCDIC codes for the
characters '[', ']', '{', and '}'.  The translate tables in hdisp, hrecv,
rjestat, and send were modified and this problem was solved.  Also, jobs
periodically disappeared.  This happened whenever the 'usr=( ... )'
specification was wrong or missing, as for example on print jobs routed
to our remote unit.  This was fixed by recompiling haspdisp with the
SFILE flag defined as '1'.
.P
We now use our RJE facility heavily and are very pleased with
its performance, although the response to 'rjestat' commands is very
slow when large files are being sent or received.  We currently use an
autodialer run by a separate command before we do haspinit, but we hope
to automate dialup and shutdown completely soon.  We have not made the
timeout modification mentioned in "I.2 Administrative Advice", but have
not noticed any problems with timeouts.  To any potential Unix RJE
installers, we say "Bon courage!"
