Re: The system cannot execute the specified program
Juergen,
Thanks for these benchmark results. One thing that is missing from the
results is the actual time to do each test and the additional time for
a customized installation of figuring out what parameters are needed
for SIZES and the time to recompile the customized NONMEM. This
customization overhead should be considered when the only criterion for
success is total runtime.
Best wishes,
Nick
Jurgen Bulitta wrote:
Dear
Nick,
Dear
All,
Please
find below the results of some benchmarking with a customized
installation
(i.e.
std installation with no. of observations and no. of compartments set
to the
smallest
values
for the respective dataset) vs. the big installation with additionally
no. of
compartments
increased
to 50 and no. of observations increased to 1001.
I
did not test any intermediately sized installations, since the results
were
reasonably
surprising,
at least to me.
The
bottom line is that, memory use did not seem to be the bottleneck, at
least not
on this
test
system. The benefit of the small installation was biggest for ADVAN1
and
ADVAN3.
However,
one will probably only notice this benefit during bootstrapping.
I
would interpret these differences in the benefit of the customized
installation
vs. the big
installation
that some parts of NONMEM are memory dependent whereas others
(especially
solving
of differential equations) are not. As long as NONMEM does not return a
memory
overflow
error and runs the job, I would not worry much about the size of the
NONMEM
installation.
Best
wishes
Juergen
Cases
1, 2, and 3:
Customized
installation: Virtual memory size: 84 MB
BIG
installation: Virtual memory size: 828 MB
(nothing
else ran on the server other than this one NONMEM VI version 6.2 run)
Case
1: Linear 1-Cmt model, ADVAN1, 15 observations, 50 subjects,
100
bootstrap resplicates with nmbs (in WFN):
Customized
installation finished 100 replicates
BIG
installation finished 61 replicates in the same time (run on a
different, but
identical computer)
-->
64% advantage for the customized installation
Case
2: Linear 2-Cmt model, ADVAN3, 15 observations, 50 subjects,
100
bootstrap resplicates with nmbs (in WFN):
Customized
installation finished 100 replicates
BIG
installation finished 76 replicates in the same time (run on a
different, but
identical computer)
-->
32% advantage for the customized installation
Case
3: Linear 3-Cmt model, ADVAN11, 15 observations, 50 subjects,
100
bootstrap resplicates with nmbs (in WFN):
Customized
installation finished 100 replicates
BIG
installation finished 89 replicates in the same time (run on a
different, but
identical computer)
-->
12% advantage for the customized installation
Case
4: Population PK model, 5 compartments, ADVAN6, run as a single job
Customized
installation: 131 min, peak memory use: 75 MB, virtual memory size: 86
MB
BIG
installation: 140 min, peak memory use: 225 MB, virtual memory size:
724 MB
-->
7% advantage for the customized installation
Case
5: same as 4, but with overloading of my laptop by running too many
programs
simultaneously
-->
7% advantage for the customized installation (i.e. exactly the same
benefit as
in case 4)
Case
6: Population PK/PD model, ADVAN6, about 40 compartments, 15
observations, run
as a single job
Customized
installation: 330.77 min
BIG
installation: 331.12 min
-->
0.11% advantage for the customized installation
Quoted reply history
From:
[email protected]
[ mailto: [email protected] ] On Behalf Of Nick Holford
Sent: Sunday, October 18, 2009 12:04 AM
To: nmusers
Subject: Re: [NMusers] The system cannot execute the specified
program
Juergen,
I'd be interested to know of your experience of how much estimation
time you
save by making project specific installations. Given that you must take
additional time to setup SIZES and recompile NONMEM for each project
(say 10-15
minutes) how much time do you save per run compared with having regular
and big
installations of NONMEM?
Nick
Jurgen Bulitta wrote:
Dear
Tom,
Dear
All,
For
those of us who prefer project specific NONMEM installations to save
estimation
time,
it
might be helpful, if ICON can provide some rough guidance on the main
contributor
to
the RAM needs of NONMEM 7.
I.e.
is it primarily the number of observations, compartments, thetas,
omegas, or
another
variable or some function of these variables?
The
following might be very cumbersome: Would it be possible to include a
switch
during
the installation which disables certain NONMEM 7 features to save RAM.
(I
recognize that this might cause a lot of unpredictable software issues.)
Dual
CPU servers currently have around 8-18 RAM slots. 2 and 4 GB RAM chips
are
reasonably
priced. Thus, on the 64-bit server side, this is probably a minor issue.
Best
wishes
Juergen
From: [email protected]
[ mailto: [email protected] ]
On Behalf Of Ludden, Thomas
Sent: Friday, October 16, 2009 3:14 PM
To: [email protected]
Cc: [email protected]
Subject: RE: [NMusers] The system cannot execute the specified
program
Pavel,
The
message you describe is usually the result of not having enough
memory available for the NONMEM 7 memory image. If you are using the
"big" version of SIZES you may need about 1.6-2 GB of
available memory for each job. For the efficient use
of NONMEM 7's new methods more variables are now stored in memory
than was the case for NONMEM VI. The memory image is larger even if
one
is not using the new methods.
When
using NONMEM 7 and the "reg" SIZES, I have been able
to run 8 jobs simultaneously on a dual processor dual core machine, but
only
3-4 jobs when using the "big" version of SIZES depending upon what
else is running. I have 3 GB of memory and the task manager indicates
a
maximum of about 7.4 GB via paging.
Tom
Thomas
M. Ludden Ph.D .
Vice-President,
Pharmacometrics
R&D
ICON
Development
Solutions
Tel:
+ 1 410 696 3040
Mob: + 1 410 258 2411
Fax: + 1 410
480 0776
Email: [email protected]
Web:
www.icondevsolutions.com
From: [email protected]
[ mailto: [email protected] ]
On Behalf Of [email protected]
Sent: Friday, October 16, 2009 2:02 PM
To: [email protected]
Cc: [email protected]
Subject: [NMusers] The system cannot execute the specified
program
> Hello NONMEM Users,
> We have a computer with 4 processors. When I try to run more
> than 3 NONMEM7 jobs, I am getting the following message:
> Starting nonmem execution ...
> The system cannot execute the specified program.
> KP9022.OUT
> newline
> 1 file(s) copied.
> Could Not Find H:\4\output
>
> The message is code and directory independent. It always
> appears when I try to run more than 3 codes. When I test
> nonmem7 using a laptop, which also has 4 processors, I can
> execute 4 programs. Also, when I had NONMEM6, I did not have
> this problem on the same machine.
> Do you familiar with this problem? Thanks.
> Pavel
>