Hi, Petr,

Not sure if you are still interested in this, but here is what I have
been puzzling over.

There are many cases where an array_t<unsigned> is deleted first, but
then copied into another array_t<unsigned>.  One possible scenario for
this is that a pointer to a bitvector is returned from a function that
generated the bitvector as a local object.  From the log file, I am
not able to tell if this true or not.

I have tried to run valgrind a few times, but not able to spot this
problem so far.  If you have a small test case for me, I am willing to
look into it a little further.

John



On 4/7/14, 11:28 PM, Petr Velan wrote:
> Hi John,
> 
> did you by any chance have a bit of time to look into this matter?
> 
> Thanks,
> Petr
> 
> 
> On Fri, Mar 21, 2014 at 1:25 PM, Petr Velan <[email protected]
> <mailto:[email protected]>> wrote:
> 
>     Hi John,
> 
>     I did the testing only with disabled mmap, no other modifications.
>     I'm now sending you an email with link to download the complete
>     log. I'm trying it on an older version from SVN revision 657.
>     Sadly, the newest is not compatible with the tested system.
> 
>     The log file is quite big 19GB, I compressed it to 2.5GB and the
>     link should be in another email shortly. I'm using
>     https://filesender.cesnet.cz/ service for the transfer.
> 
>     Let me know if there is anything more that I can do.
> 
>     Best,
>     Petr
> 
> 
> 
> 
>     On Thu, Mar 20, 2014 at 5:21 PM, K. John Wu <[email protected]
>     <mailto:[email protected]>> wrote:
> 
>         Hi, Petr,
> 
>         FastBit is only tracking the memory usage of a few object
>         types such
>         as ibis::array_t and ibis::buffer.  Since you are using a
>         compression
>         layer, I presume that you are also allocating considerable
>         amounts of
>         memory using malloc or new directly.  This would mean FastBit
>         is only
>         tracking a part of the memory uses.  If you turn up the
>         verboseness of
>         the FastBit code, say, ibis::util::setVerboseLevel(10), and
>         ship me
>         the log messages, I can take a look and see if I can spot anything
>         more useful to suggest.
> 
>         John
> 
> 
>         On 3/20/14, 6:35 AM, Petr Velan wrote:
>         > Hi John,
>         >
>         > I'm currently working on FastBit support for compressed
>         files. To
>         > simplify things a bit for a start, I have disabled the MMAP
>         support
>         > and forced FastBit to use open, read, write and seek
>         functions. Then I
>         > hooked into these functions and provided a compression
>         layer. The
>         > problem is that just disabling the MMAP causes FastBit to
>         use so much
>         > memory that OOM killer terminates the process for some
>         aggregation
>         > queries.
>         >
>         > Is there anything I can do to tell FastBit to limit the
>         amount of
>         > memory it can use? I thought that only half of the available
>         memory
>         > will be used by default. I've tested this issue on latest
>         SVN revision.
>         >
>         > Thank you in advance for any help,
>         > Petr Velan
>         >
>         >
>         > _______________________________________________
>         > FastBit-users mailing list
>         > [email protected]
>         <mailto:[email protected]>
>         > https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
>         >
>         _______________________________________________
>         FastBit-users mailing list
>         [email protected] <mailto:[email protected]>
>         https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
> 
> 
> 
> 
> 
> _______________________________________________
> FastBit-users mailing list
> [email protected]
> https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
> 
_______________________________________________
FastBit-users mailing list
[email protected]
https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users

Reply via email to