SMP issue, OSR 507, Office Portfolio
Roger Cornelius
several years as we have upgraded UNIX/OpenServer from 3.2v4.2 to the
current OSR 5.0.7. Starting with about 5.0.2 (I think) OP would no
longer run if we had the SMP software installed. OP would fail with the
message:
MV wiockeys() failed! Can't register hotkeys
This is actually the message I get with 5.0.7 but it is the same or
similar as previous OS versions. Since sar reports didn't show any
performance bottleneck, I just uninstalled SMP and we ran without it and
OP works fine in that case. We're now in the process of upgrading an
aging server to a new one, which, though we didn't order it, was shipped
with a second processor installed. I tried installing SMP and, not
surprisingly, OP won't run. I know that OP hasn't been supported for
years, but I'd love to be able to use it and the second processor at the
same time. Is there any chance of making it work? On a side note, the
new server has XEON processors. If I can't run OP with SMP, will I be
able to run it with one XEON processor with the osr507up Hyper-Threading
update installed?
I've run truss against the OP pmiew.rts binary. It forks a new process
which ultimately dies with a segmentation fault. I can provide the
trace if needed.
Thanks for any help.
--
Roger Cornelius rac...@tenzing.org
Bela Lubkin
OSR5 `truss` is imperfect; one of its faults is that it sometimes causes
a process to SIGSEGV when it would have been fine when run standalone.
It's possible the process was going to SIGSEGV by itself, but that's
probably an observation artifact.
My desktop has a Pentium 4 "2.40C GHz", 800FSB, that supports
hyperthreading. I installed OP on it and it fails with the "wiockeys"
message on some runs. I actually see three different behaviors:
sometimes it hangs, sometimes it fails with "wiockeys", sometimes it
starts successfully (with a "Caldaemon process has stopped due to a
fatal error\nF1 to clear reminder" complaint). BTW when I say
"installed" I actually mean "copied files from a 3.2v4.2 system that
once ran OP" -- I don't even know if it works correctly on the original
system, and I don't know anything about running, installing or
administrating the beast myself.
When I `truss`, it usually does the third thing (seems to work).
If I disable a CPU (`cpuonoff -i 2`), it usually seems to work (with or
without `truss`).
I'm doing the `truss` in the `op` shell script,
runapp truss -o /tmp/op.truss $OALIB/mvw/pmview.rts -R "$OALIB/oadaemon"
instead of:
runapp $OALIB/mvw/pmview.rts -R "$OALIB/oadaemon"
The variant behavior between truss/no-truss and SMP/disabled strongly
implies that this is a timing problem. It's probably forking a child,
then assuming some events will occur in a certain sequence which isn't
true on an SMP system.
Since you know how to install & use OP, you're in a better position to
test this...
Pathetic idea, but, what if you `cpuonoff -i 2` (etc.), disable all your
extra processors; then start OP; then reenable them. Does OP continue
to work? Obviously this isn't a viable way to run a system, I'm just
trying to figure out if the problem is pervasive, or if there's just one
small hump to get over.
>Bela<
Bob Bailin
"Bela Lubkin" <be...@sco.com> wrote in message
news:20030813023...@sco.com...
Would lockpid achieve the same results? (Are child processes locked to
the same CPU as the parent?)
Bob
Roger Cornelius
OK. The 5.0.7 truss seems to have matured a bit from the scotruss I
got somewhere long ago though.
I'm also doing the truss in the op script, but in the runapp function
itself. Essentially the same as what you're doing.
This is an IBM x345 server with two XEON 2.67ghz processors. With the
second cpu active, I am unable to get op to start successfully at all
using truss or no. After more testing, op does hang occasionally
instead of printing the "wiockeys" message. The majority of times I
get the message though.
If I switch the second cpu off, I no longer get the "wiockeys"
message. Instead, op just clears the screen and then exits with a 0
status. If I capture the output, I see it actually spits out a bunch
of terminal control codes. This happens consistently - I ran a loop
of 50 iterations of "op; sleep 3" and each one behaved this way.
Still no successful starts.
Do you have any other suggestions?
Bela Lubkin
> "Bela Lubkin" <be...@sco.com> wrote in message
> news:20030813023...@sco.com...
> > Pathetic idea, but, what if you `cpuonoff -i 2` (etc.), disable all your
> > extra processors; then start OP; then reenable them. Does OP continue
> > to work? Obviously this isn't a viable way to run a system, I'm just
> > trying to figure out if the problem is pervasive, or if there's just one
> > small hump to get over.
>
> Would lockpid achieve the same results? (Are child processes locked to
> the same CPU as the parent?)
I think only the one process is locked; I'm pretty sure I've tested that
both fork() and exec() unlock it. Also, the `lockpid` command wouldn't
really help for this because the failure is immediate, and `lockpid`
operates on a PID. You would want a `lockcmd` command that started the
process and immediately locked it.
Doesn't matter, it's just for test purposes. And Roger says it didn't
help enough on his system.
>Bela<