Date: Wed, 5 Sep 2012 11:47:03 -0400 (EDT)
From: Jun-Hui Zhao 
To: Peter Teuben 
Subject: Re: miriad bugs


The problem shows in uvclose_c with my SMALOD (sma_mirRead.c).
Two large datasets were produced from SMA testing runs:

1) integration 1.3 sec
   # of int    13983
   # of bsln   28
   # of chan   6145
   # of rx     1
   size        20 Gbytes

2) integration 1.3 sec
   # of int    48536
   # of bsln   10
   # of chan   6145
   # of rx     2
   size        50 Gbytes

   I have no problems to handle the dataset 1 (20GB of 13983).
The segfault occurs at the end of loading (uvclose)  the
dataset 2 (50GB of 48536 ints) that was produced in the mid of
August before my week off.

   Jun-Hui


On Wed, 5 Sep 2012, Peter Teuben wrote:

> where does this problem show up?   I have a very small
> test routine which might be able to reproduce this, the
> $MIR/install/testmirlib.c  has very simple code to create
> some"fake" vis data, but a lot
>
>
> cd $MIR/install
> set lfs=(-D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE)
> gcc -g $lfs -I$MIRINC -I$MIRSUBS -o testmirlib testmirlib.c
> $MIRLIB/libmir.a -lm
>
> then
>
> ./testmirlib u 4 4 100000
>
> would write 100,000 records, but only 4 channel and 4 wideband channel
> data. So, small
> dataset in terms of channels. You can go as big as you want.
>
>
> The mir.bigbench script uses uvrandom and writes gigantic data. The
> default is actually 525,000
> data.
>
> If you can tell me what the size parameters are, i might be able to see
> what's going on. I haven't
> tried dataset over 16 GB i must admit. I'll try something huge with
> mir.bigbench now.
>
> peter
>
> On 09/05/2012 10:53 AM, Jun-Hui Zhao wrote:
> > Hi Peter,
> >     Thanks.
> >     I got stucked at a problem. The 4.3.4 miriad appears to be able
> > to handle large size files of SMA data (> 50 Gbytes) but I got
> > segfault from uvclose_c when the data involves a large number of

....