[Distrs] Salutations

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[Distrs] Salutations

NikosAlexandris
Congratulations for this new list and best of success.  Where can one start
testing, just one step at a time?

Thank you, Nikos
_______________________________________________
distRS mailing list
[hidden email]
http://lists.osgeo.org/cgi-bin/mailman/listinfo/distrs
Reply | Threaded
Open this post in threaded view
|

Re: [Distrs] Salutations

Yann Chemin-3
Hi Nikos,

There are two sets of distributed suites, one using OpenMP (complete) and a small one using OpenCL (incomplete but few simple modules are there).

In any case I suggest you take any of the directory inside prog/ or progcl/ that you are most familiar with (i.e. prog_NDVI is always an easy take) and compile it in your computer (make clean; make). After that, use any input required (for NDVI, Red and NIR) and run the module. The distribution of processing is done within the dataset processed, i.e. all pixels are loaded in RAM, then each receives pixels as available and process the image required.

for OpenMP (prog/ directory) distribution, I suggest you use a biggish image (1Gb per input band) to see a significant change with 4/8 core/threads. It scales well, you can try bigger and more cores/threads.

Cheers,
Yann

PS: if your CPU is INTEL, go in the BIOS and enable HyperThreading, it will double you number of cores, works very well for such type of processing distribution.



On 20 November 2013 15:44, Nikos Alexandris <[hidden email]> wrote:
Congratulations for this new list and best of success.  Where can one start
testing, just one step at a time?

Thank you, Nikos
_______________________________________________
distRS mailing list
[hidden email]
http://lists.osgeo.org/cgi-bin/mailman/listinfo/distrs



--
----

_______________________________________________
distRS mailing list
[hidden email]
http://lists.osgeo.org/cgi-bin/mailman/listinfo/distrs
Reply | Threaded
Open this post in threaded view
|

Re: [Distrs] Salutations

NikosAlexandris
Yann Chemin wrote:

> Hi Nikos,
>
> There are two sets of distributed suites, one using OpenMP (complete) and a
> small one using OpenCL (incomplete but few simple modules are there).

Thank you Yann.  I will try to have a look at his next week.  Earlier it is
impossible.
 
> In any case I suggest you take any of the directory inside prog/ or progcl/
> that you are most familiar with (i.e. prog_NDVI is always an easy take) and
> compile it in your computer (make clean; make). After that, use any input
> required (for NDVI, Red and NIR) and run the module. The distribution of
> processing is done within the dataset processed, i.e. all pixels are loaded
> in RAM, then each receives pixels as available and process the image
> required.

Sounds easy :-p

> for OpenMP (prog/ directory) distribution, I suggest you use a biggish
> image (1Gb per input band) to see a significant change with 4/8
> core/threads. It scales well, you can try bigger and more cores/threads.

Will do so.
 
> Cheers,
> Yann
>
> PS: if your CPU is INTEL, go in the BIOS and enable HyperThreading, it will
> double you number of cores, works very well for such type of processing
> distribution.

Working with a Xeon E5-2630 here. Funtoo sees 12 cores, so I guess it's
allready at max.

Nikos
_______________________________________________
distRS mailing list
[hidden email]
http://lists.osgeo.org/cgi-bin/mailman/listinfo/distrs
Reply | Threaded
Open this post in threaded view
|

Re: [Distrs] Salutations

Yann Chemin-3
Hi Nikos,

yes 12 cores, is most probably 6 cores x2 HyperThreads, = 12 "virtual cores".

You are ready to get started with the code!
Cheers,
Yann


On 27 November 2013 06:17, Nikos Alexandris <[hidden email]> wrote:
Yann Chemin wrote:

> Hi Nikos,
>
> There are two sets of distributed suites, one using OpenMP (complete) and a
> small one using OpenCL (incomplete but few simple modules are there).

Thank you Yann.  I will try to have a look at his next week.  Earlier it is
impossible.

> In any case I suggest you take any of the directory inside prog/ or progcl/
> that you are most familiar with (i.e. prog_NDVI is always an easy take) and
> compile it in your computer (make clean; make). After that, use any input
> required (for NDVI, Red and NIR) and run the module. The distribution of
> processing is done within the dataset processed, i.e. all pixels are loaded
> in RAM, then each receives pixels as available and process the image
> required.

Sounds easy :-p

> for OpenMP (prog/ directory) distribution, I suggest you use a biggish
> image (1Gb per input band) to see a significant change with 4/8
> core/threads. It scales well, you can try bigger and more cores/threads.

Will do so.

> Cheers,
> Yann
>
> PS: if your CPU is INTEL, go in the BIOS and enable HyperThreading, it will
> double you number of cores, works very well for such type of processing
> distribution.

Working with a Xeon E5-2630 here. Funtoo sees 12 cores, so I guess it's
allready at max.

Nikos



--
----

_______________________________________________
distRS mailing list
[hidden email]
http://lists.osgeo.org/cgi-bin/mailman/listinfo/distrs