Borrar filtros
Borrar filtros

Where can I buy Server capacity?

3 visualizaciones (últimos 30 días)
Chewy
Chewy el 10 de En. de 2016
Comentada: Walter Roberson el 11 de En. de 2016
I have an image processing script which takes nearly 3 weeks to be computated on my notebook, but I need the results earlier. Is there any provider of server capacity, where I can upload my files and the script is run on bigger machines?

Respuesta aceptada

Walter Roberson
Walter Roberson el 10 de En. de 2016
A server farm will not necessarily help much -- unless your code can be parallelized effectively, which is not always the case.
Changing the algorithm or writing in a more vectorized form can often help a lot more than running on a server that might only be 4/3 times as fast per CPU.
When you profile the code, where is it spending its time?
  2 comentarios
Chewy
Chewy el 11 de En. de 2016
Editada: Chewy el 11 de En. de 2016
I do a template matching with 80 to 110 images. Before calculating the similarity I do calculate the nearly 12000 permutations per image and save them in a 400 MB big struct for each image. More than 60 % of the time matlab spend on loading these structs. My 12GB RAM are overloaded, so that the pagefile is nearly 30 GB big. I think, there are potencials to optimise the data structure, but I have no clue how to do this. A full parallelisation is hard to implement with my approach. Unfortunatly I don´t know, if I´m allowed to present here the complete code. So I think, I will leave it like it is and wait for the results. But thanks!
Walter Roberson
Walter Roberson el 11 de En. de 2016
In the situation where you are swapping, running on a server with a lot of memory might be worthwhile.

Iniciar sesión para comentar.

Más respuestas (1)

jgg
jgg el 10 de En. de 2016
I think this page provides you with a few options. The Amazon EC2 option is probably what you want since it bills hourly.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by