VAXELN is new to me, but I am reasonably at home in
the internals of RSX11M,
MTS and OS 360. I was called in by the application programmer when he
encountered a somewhat deep problem.
One or two times a month in the 3rd of three RT1000 running V4.0 of VAXELN
the two communications tasks with the microVax 3300 running VMS 5.2 stop.
One with access violation, reason mask 05 and one with access violation or
with a kernel stack not valid exception. These tasks are identical and each
serves its own sorting unit. The VAXELN systems use the DDCMP_V2 module. The
systems run continuously. The stop address is for both tasks always in the
DAP module (from june 7 1989) at 1FFC from the start
It looks as if the FP register contains 0 during the last CLRB -4C4(FP)
intruction.I conclude this from the virtual address =FFFFFB3C message..
The system stopped at the same location during startup earlier in the
project each time when the 3rd realtime vax was started, but then:maximum
links was 32 and maximum circuits was16. This error disappeared when these
quantities were doubled to maximum links=64 and maximum circuits=32. The
earlier values were ok during the years when only two realtime vaxes were
active.and with 2 realtime vaxes these crashes were not noticed.
The realtime vaxes all have the remote debugger. The installation is 140 km
from my home.
It looks as if register information is hard to get at. Suggestions?
Has anyone encountered a similar error? Who has suggestions?.
When this gets highly detailed contact off list might be best.
[I'd trim the above if it didn't mean cutting something out
that should stay]
First off, In my experience with VAXELN 4.x you can see these
sort of failures due to excess network traffic. We solved that
problem by isolating the VAXELN system on a local network and
put two network interfaces in the VMS system and set it up as
a DECnet router (because we needed the VMS system to talk to
both networks).
If there's not a problem with increasing maximum links and
maximum circuits, and that solves the problem, then you can
just leave those values doubled.
--
Eric Dittman
dittman(a)dittman.net
Check out the DEC Enthusiasts Club at
http://www.dittman.net/