[mpich-discuss] Running an mpi program that needs to access /dev/mem

[Jeff Hammond]

I do not own a Rasberry Pi.  Is this thing self-hosted, hence you are
driving the Hydra download+build from a wimpy ARM core?  Or is your Pi
attached to a real computer?  In case it wasn't obvious from the shell
prompt ("Jeffs-MacBook-Pro:tmp jhammond"), I was timing on a Macbook
Pro, which happens to have a 2.5 GHz Intel Core i7 quad-core
processor.  When I time on the oldest processor to which I have access
(PPC970MP 2.5 GHz circa 2007), the whole process takes twice as long
(77 seconds).  In both cases, I built in /tmp, which on Linux is a
ramdisk (http://en.wikipedia.org/wiki/Tmpfs).

If building Hydra is a bottleneck because the Pi is too slow for basic
operations like download a file, unpack it and run gcc on its
contents, then you should go back to your very first statement ("I am
trying to use two raspberry-pi to sample and then process some data")
and consider whether or not what you're doing makes sense.  If a
computer can't compile very simple software like Hydra in a reasonable
amount of time, I fail to see how it can be used to process data
quickly.  While configure isn't necessarily a good model for data
analysis, I imagine that your time is best spent implementing and
running your data analysis on a computer with more substantial
processing power.

I guess I should also point out that you should be able to cross
compile Hydra if you have the appropriate cross-compiler toolchain,
hence you can download and build Hydra on an x86 laptop and then
merely copy the resulting binary over to your toy processor.

Best,

Jeff

On Sun, Jun 16, 2013 at 12:26 PM, Lee, Eibhlin
<eibhlin.lee10 at imperial.ac.uk> wrote:
> Unfortunately my past experience with hydra is VERY different.
> I'm baffled as to how you managed to do it so quickly. Surely you have to download the file from mpich.org then unzip (which takes at least 5 mins) then configure (up to an hour) make (a long time as well) and finally make install (another chance in which you can go have a cuppa). And then get the image and put it onto another device which first needs formatting; for me that takes anywhere between 10 minutes and an hour.
> What did you do so that it took so little time?
> Eibhlin
> ________________________________________
> From: discuss-bounces at mpich.org [discuss-bounces at mpich.org] on behalf of Jeff Hammond [jhammond at alcf.anl.gov]
> Sent: 16 June 2013 18:19
> To: discuss at mpich.org
> Subject: Re: [mpich-discuss] Running an mpi program that needs to access        /dev/mem
>
>> I did not want to spend a whole day installing hydra instead of smpd if it would not work.
>
> It took me 41 seconds to install Hydra ex nihilo and I didn't even use
> "make -j".  The argument that installation time is a bottleneck is
> invalid.  smpd is not supported and you shouldn't use it.  End of
> discussion.
>
> Jeff
>
> <snip>
> ALL DONE
> Sun Jun 16 12:16:12 CDT 2013
>
> real    0m40.912s
> user    0m17.405s
> sys     0m11.028s
>
> Jeffs-MacBook-Pro:tmp jhammond$ cat auto
> #!/bin/bash
> date
> wget http://www.mpich.org/static/downloads/3.0.4/hydra-3.0.4.tar.gz && \
> tar -xzf hydra-3.0.4.tar.gz && \
> cd hydra-3.0.4 && \
> mkdir build && \
> cd build && \
> ../configure CC=gcc --prefix=/tmp/hydra-install && \
> make && \
> make install && \
> make check
> echo "ALL DONE"
> date
>
> --
> Jeff Hammond
> Argonne Leadership Computing Facility
> University of Chicago Computation Institute
> jhammond at alcf.anl.gov / (630) 252-5381
> http://www.linkedin.com/in/jeffhammond
> https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond
> ALCF docs: http://www.alcf.anl.gov/user-guides
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss



-- 
Jeff Hammond
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
http://www.linkedin.com/in/jeffhammond
https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond
ALCF docs: http://www.alcf.anl.gov/user-guides