What I mean is that negative memory reported by Windows is not responsible for the crash in all OS for high orders
On August 12, 2015 6:02:52 PM CEST, Johannes Pfeifer jpfeifer@gmx.de wrote:
It's the same mod-file, Approx_10.mod, just used in the two programs with order=3. As far as I understood, they rely on the same underlying routines.
Johannes Pfeifer Friesenwall 104 50672 Köln Mobil: +49-(0)170-6936820 jpfeifer@gmx.de
-----Ursprüngliche Nachricht----- Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Michel Juillard Gesendet: Mittwoch, 12. August 2015 18:01 An: List for Dynare developers Betreff: Re: [DynareDev] Dynare++ Memory
OK, but this is a problem different from the one encountered with Approx_10.mod
Johannes Pfeifer writes:
Dynare++ always diagnoses 0 memory and k_order_pert.mexw64 in Dynare
unstable always negative memory.
-----Ursprüngliche Nachricht----- Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Michel Juillard Gesendet: Mittwoch, 12. August 2015 17:29 An: List for Dynare developers Betreff: Re: [DynareDev] Dynare++ Memory
Are you saying that under Windows, dynare++ always diagnoses negative
memory?
Johannes Pfeifer writes:
Sébastien also indicated some time ago that this message "just" triggers a different algorithm. But this is still a very undesirable
behavior. I get this exact same message in the jnl file that is created when I use this mod-file with Dynare under Matlab at order 3
with the k_order_pert.mex64. In this case, at order=3 memory should not be an issue. But the mex-file diagnoses negative memory and immediately switches to the different algorithm from the start.
Johannes Pfeifer Friesenwall 104 50672 Köln Mobil: +49-(0)170-6936820 jpfeifer@gmx.de
-----Ursprüngliche Nachricht----- Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Michel Juillard Gesendet: Mittwoch, 12. August 2015 15:28 An: List for Dynare developers Betreff: Re: [DynareDev] Dynare++ Memory
Looking again the the journal, I believe now that the message "run out of memory" is not an error message but an information that triggers a change in the trade-off between speed and memory (i.e. decreasing number of the threads)
I don't think that dynare++ catches memory allocation failure and just crashes at that point.
Michel Juillard writes:
Johannes,
do you know for which order this example is working on your machine
and for which order it doesn't?
When it is working does the jnl make sense?
It may be that 8GB is not enough to even store the derivatives of the model before starting computations but that for some reason the
program doesn't crash until much later.
Best
Michel
Johannes Pfeifer writes:
Hi Michel,
I ran dynare++ on the mentioned mod-file again under Windows++ and
used the system monitoring tools for monitoring memory usage. The picture is attached. The blue line shows the percent of memory
used
of my 8GB RAM on my Laptop from the start until the crash. As you can see, there is not much movement. I do not get anywhere close
to
the 8GB of memory used for the task that you report for Linux. Rather, straight from the beginning the jnl-file reports 0 available memory and "out of memory". It simply seems as if Dynare++ is hardly using any memory straight from the start. This Dynare++ does not seem to be just a problem of measuring memory usage on Windows.
Best,
Johannes
Johannes Pfeifer
Department of Economics
University of Mannheim
L7, 3-5, Room 242
68131 Mannheim
Germany
+49 (0)621 181-3430
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
-- Michel Juillard _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev