[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: sunset downstream sites may feedfrom profhorn.meteor.wisc.edu

Pete,

Sorry to hear replacing the memory and the other things you've tried
haven't fixed the problem.  The dbx traceback you sent showing the bus
seems to indicate an alignment problem, as if something is being
stored at an address that is not properly aligned for the type of data
that is stored there, for example trying to store a 32-bit integer at
an odd byte address.

I can't remember seeing anything quite like that, and I couldn't
reproduce the problem on an SGI/IRIX 6.5 platform here.

Your experiment with changing queue sizes to show that 67 products
works but 68 products doesn't leads me to believe you might be able to
explicitly set the number of products to a larger number using the
"-S" option to pqcreate.  While you're at it, you should probably be
using the "-c" (clobber) option as well, so you don't have to manually
delete the queue each time before you create a new one.

pqcreate just divides the queue size by 4096 to get the number of
product slots to use, but you can specify a different number with the
-S option, something like:

  pqcreate -c -q /cool.pretty/ldm/ldm.pq -v -s 25000000 -S 6101

for example to make the queue have 6101 product slots instead of 6103.
If you played around with this, you might find a value that worked
with a large queue and there might be a pattern to the bus errors that
depends on the number of product slots.

This is pure speculation since I can't reproduce the problem, but
maybe you are compiling with a compiler flag or optimization level
that changes the alignment restrictions.  For example, if you set the
highest level of optimization when compiling, maybe that requires
strict alignment, whereas if you don't specify optimization but
instead use the debugging flag "-g", looser alignment works.

I'm afraid I'll have to wait until Monday to pursue this, but a little
more information might help:

 - Do you have the CFLAGS environment variable set when you build the
   LDM?  If so, what value?

 - Is this the first time you've tried LDM 5.1.2 on this SGI/IRIX
   platform (sunset)?  If so, what version were you running with
   successfully before?

 - What kind of platform is profhorn?  Are you using LDM 5.1.2 on it?

You may have found a platform-specific bug in LDM 5.1.2, but until we
can reproduce it, we'll have trouble fixing it ...

--Russ